yslan commited on
Commit
72209bd
·
verified ·
1 Parent(s): 25c87d4

Upload folder using huggingface_hub

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +8 -0
  2. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/ismb_sbatch_14b_8node_288x512_fsdp_raymaptokens_nextstage_recammaster_pair1x49_eb32_lr2e5.sh +170 -0
  3. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml +77 -0
  4. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/torch_profiler/rank0.1780383301809004862.pt.trace.json +3 -0
  5. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/debug-internal.log +95 -0
  6. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/debug.log +24 -0
  7. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/files/config.yaml +700 -0
  8. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/files/output.log +212 -0
  9. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/files/requirements.txt +208 -0
  10. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/files/wandb-metadata.json +235 -0
  11. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/files/wandb-summary.json +1 -0
  12. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/logs/debug-internal.log +95 -0
  13. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/logs/debug.log +24 -0
  14. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/run-3241szoo.wandb +3 -0
  15. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/ismb_sbatch_14b_8node_288x512_fsdp_raymaptokens_nextstage_recammaster_pair1x49_eb32_lr2e5.sh +170 -0
  16. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml +77 -0
  17. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/torch_profiler/rank0.1780383283598487142.pt.trace.json +3 -0
  18. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/debug-internal.log +97 -0
  19. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/debug.log +24 -0
  20. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/files/config.yaml +700 -0
  21. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/files/output.log +234 -0
  22. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/files/requirements.txt +208 -0
  23. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/files/wandb-metadata.json +235 -0
  24. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/files/wandb-summary.json +1 -0
  25. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/logs/debug-internal.log +97 -0
  26. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/logs/debug.log +24 -0
  27. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/run-ixxjov67.wandb +3 -0
  28. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/ismb_sbatch_14b_8node_288x512_fsdp_raymaptokens_nextstage_recammaster_pair1x49_eb32_lr2e5.sh +170 -0
  29. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml +77 -0
  30. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/torch_profiler/rank0.1780383278694422947.pt.trace.json +3 -0
  31. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/debug-internal.log +87 -0
  32. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/debug.log +24 -0
  33. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/files/config.yaml +700 -0
  34. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/files/output.log +226 -0
  35. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/files/requirements.txt +208 -0
  36. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/files/wandb-metadata.json +235 -0
  37. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/files/wandb-summary.json +1 -0
  38. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/logs/debug-internal.log +87 -0
  39. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/logs/debug.log +24 -0
  40. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/run-nn58xrpu.wandb +3 -0
  41. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/ismb_sbatch_14b_8node_288x512_fsdp_raymaptokens_nextstage_recammaster_pair1x49_eb32_lr2e5.sh +170 -0
  42. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml +77 -0
  43. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/torch_profiler/rank0.1780391935713307947.pt.trace.json +3 -0
  44. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/debug-internal.log +89 -0
  45. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/debug.log +24 -0
  46. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/run-20260602_091502-iwxrpu9h/files/config.yaml +700 -0
  47. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/run-20260602_091502-iwxrpu9h/files/output.log +248 -0
  48. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/run-20260602_091502-iwxrpu9h/files/requirements.txt +208 -0
  49. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/run-20260602_091502-iwxrpu9h/files/wandb-metadata.json +235 -0
  50. ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/run-20260602_091502-iwxrpu9h/files/wandb-summary.json +1 -0
.gitattributes CHANGED
@@ -33,3 +33,11 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/torch_profiler/rank0.1780383301809004862.pt.trace.json filter=lfs diff=lfs merge=lfs -text
37
+ ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/run-3241szoo.wandb filter=lfs diff=lfs merge=lfs -text
38
+ ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/torch_profiler/rank0.1780383283598487142.pt.trace.json filter=lfs diff=lfs merge=lfs -text
39
+ ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/run-ixxjov67.wandb filter=lfs diff=lfs merge=lfs -text
40
+ ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/torch_profiler/rank0.1780383278694422947.pt.trace.json filter=lfs diff=lfs merge=lfs -text
41
+ ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/run-nn58xrpu.wandb filter=lfs diff=lfs merge=lfs -text
42
+ ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/torch_profiler/rank0.1780391935713307947.pt.trace.json filter=lfs diff=lfs merge=lfs -text
43
+ ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/run-20260602_091502-iwxrpu9h/run-iwxrpu9h.wandb filter=lfs diff=lfs merge=lfs -text
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/ismb_sbatch_14b_8node_288x512_fsdp_raymaptokens_nextstage_recammaster_pair1x49_eb32_lr2e5.sh ADDED
@@ -0,0 +1,170 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+ #SBATCH --job-name=14b_8node_288x512_pair1x49_next_eb32_lr2e5
3
+ #SBATCH --nodes=8
4
+ #SBATCH --gpus=32
5
+ #SBATCH --gpus-per-node=4
6
+ #SBATCH --ntasks-per-node=1
7
+ #SBATCH --time=24:00:00
8
+ #SBATCH --output=out/exp_%j_%x.out
9
+
10
+ # Track4D-360 14B FSDP 288x512 next-stage finetune.
11
+ #
12
+ # Main changes from the 288p backendmix OOM-debug launcher:
13
+ # - new DATA_TAG / output directory;
14
+ # - warm-start from the latest existing 288p backendmix checkpoint;
15
+ # - remove most static datasets from DATA_CONFIG;
16
+ # - add ReCamMaster CoWTracker + square512 source49 + compact pair1x49
17
+ # VGGT-Omega labels, q1-only by dataset construction.
18
+ #
19
+ # Submit from repo root on Isambard:
20
+ # sbatch bash_scripts/track4d_360/ismb/14b/sbatch/ismb_sbatch_14b_8node_288x512_fsdp_raymaptokens_nextstage_recammaster_pair1x49_eb32_lr2e5.sh
21
+
22
+ export NNODES=8
23
+ export MODEL_SIZE=14b
24
+ export CONDA_ENV="${CONDA_ENV:-th29py312}"
25
+ export HEIGHT=288
26
+ export WIDTH=512
27
+ export NUM_FRAMES=49
28
+ export main_process_port="${main_process_port:-29595}"
29
+
30
+ export WAN_T2V14B_DIR="${WAN_T2V14B_DIR:-${SLURM_SUBMIT_DIR}/weights/Wan2.1-T2V-14B}"
31
+ export VXF_INIT="${VXF_INIT:-${WAN_T2V14B_DIR}/diffusion_pytorch_model*.safetensors}"
32
+ export MODEL_ID_PREFIX="${MODEL_ID_PREFIX:-Wan-AI/Wan2.1-T2V-14B}"
33
+ export TOKENIZER_PATH="${TOKENIZER_PATH:-${WAN_T2V14B_DIR}/google/umt5-xxl}"
34
+ export T5_PATH="${T5_PATH:-${WAN_T2V14B_DIR}/models_t5_umt5-xxl-enc-bf16.pth}"
35
+ export VAE_PATH="${VAE_PATH:-${WAN_T2V14B_DIR}/Wan2.1_VAE.pth}"
36
+
37
+ # Prefer the current 288p backendmix run as the next-stage warm start. Fall
38
+ # back to the 384p backendmix/raymap-token parents only when no 288p checkpoint exists.
39
+ export BACKENDMIX_288_PARENT_DIR="${BACKENDMIX_288_PARENT_DIR:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_warmstart384_backendmix_q1to4_s1248_jitter_norecam_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_warmstart384_backendmix_q1to4_s1248_jitter_norecam_eb32_288x512_cond_dropout}"
40
+ export BACKENDMIX_384_PARENT_DIR="${BACKENDMIX_384_PARENT_DIR:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_resume_latest_backendmix_q1to4_s1248_jitter_norecam_eb32_384x672_cond_dropout_384x672x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_resume_latest_backendmix_q1to4_s1248_jitter_norecam_eb32_384x672_cond_dropout}"
41
+ export RAYMAP_384_PARENT_DIR="${RAYMAP_384_PARENT_DIR:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_real_synth_syn4d_recam_syncam_softsplat_raymaptokens_t2v_init_scratch_eb32_cond_dropout_384x672x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_real_synth_syn4d_recam_syncam_softsplat_raymaptokens_t2v_init_scratch_eb32_cond_dropout}"
42
+ # Warm-start from the 288p backendmix run's step-9000 checkpoint — the requested
43
+ # next-stage starting point. (FSDP1/FSDP2 read it identically: warm-start loads
44
+ # the safetensors BEFORE accelerator.prepare, on the unwrapped model, so it is
45
+ # distributed-backend-independent.) Override WARM_START=... to pick another;
46
+ # falls back to latest backendmix / 384p parents if step-9000 is absent.
47
+ if [ -z "${WARM_START:-}" ]; then
48
+ if [ -f "${BACKENDMIX_288_PARENT_DIR}/step-9000.safetensors" ]; then
49
+ export WARM_START="${BACKENDMIX_288_PARENT_DIR}/step-9000.safetensors"
50
+ else
51
+ latest="$(find "${BACKENDMIX_288_PARENT_DIR}" -maxdepth 1 -name 'step-*.safetensors' 2>/dev/null | sort -V | tail -1)"
52
+ if [ -z "${latest}" ]; then
53
+ latest="$(find "${BACKENDMIX_384_PARENT_DIR}" -maxdepth 1 -name 'step-*.safetensors' 2>/dev/null | sort -V | tail -1)"
54
+ fi
55
+ if [ -z "${latest}" ]; then
56
+ latest="$(find "${RAYMAP_384_PARENT_DIR}" -maxdepth 1 -name 'step-*.safetensors' 2>/dev/null | sort -V | tail -1)"
57
+ fi
58
+ export WARM_START="${latest:-${RAYMAP_384_PARENT_DIR}/step-3500.safetensors}"
59
+ fi
60
+ fi
61
+
62
+ if ! compgen -G "${VXF_INIT}" >/dev/null; then
63
+ echo "ERROR: VXF_INIT not found: ${VXF_INIT}" >&2
64
+ exit 1
65
+ fi
66
+ [ -d "${TOKENIZER_PATH}" ] || { echo "ERROR: TOKENIZER_PATH not found: ${TOKENIZER_PATH}" >&2; exit 1; }
67
+ [ -f "${T5_PATH}" ] || { echo "ERROR: T5_PATH not found: ${T5_PATH}" >&2; exit 1; }
68
+ [ -f "${VAE_PATH}" ] || { echo "ERROR: VAE_PATH not found: ${VAE_PATH}" >&2; exit 1; }
69
+
70
+ export TARGET_EFFECTIVE_BATCH=32
71
+ export gradient_accumulation_steps=1
72
+ export learning_rate="${learning_rate:-2e-5}"
73
+ export save_steps="${save_steps:-500}"
74
+ export dataset_num_workers="${dataset_num_workers:-7}"
75
+ export TRACK4D360_STOP_AFTER_NEXT_SAVE="${TRACK4D360_STOP_AFTER_NEXT_SAVE:-0}"
76
+
77
+ export dit_finetune_modules="${dit_finetune_modules:-self_attn}"
78
+ export DIT_FINETUNE_TAG="${DIT_FINETUNE_TAG:-attn}"
79
+ export use_plucker_raymap_tokens=1
80
+ export train_plucker_camera_control=0
81
+ export PLUCKER_WANDB_TAG="${PLUCKER_WANDB_TAG:-plucker_raymap_tokens}"
82
+
83
+ export DATA_CONFIG="${DATA_CONFIG:-${SLURM_SUBMIT_DIR}/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml}"
84
+ export DATA_TAG="${DATA_TAG:-mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512}"
85
+
86
+ export RECAMMASTER_SOURCE49_ROOT="${RECAMMASTER_SOURCE49_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_square512_source49}"
87
+ export RECAMMASTER_PAIR1X49_LABEL_ROOT="${RECAMMASTER_PAIR1X49_LABEL_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_pair1x49_square512_scale_labels}"
88
+ export RECAMMASTER_COW_ROOT="${RECAMMASTER_COW_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_VGGT_Omega_CoWTracker_SourceTracks_all}"
89
+ export TRACK4D360_RECAM_TRACK_SAMPLE_MODE="${TRACK4D360_RECAM_TRACK_SAMPLE_MODE:-grid}"
90
+ export TRACK4D360_RECAM_TRACK_STRIDE="${TRACK4D360_RECAM_TRACK_STRIDE:-2}"
91
+ export TRACK4D360_RECAM_TRACK_SAMPLES="${TRACK4D360_RECAM_TRACK_SAMPLES:-0}"
92
+ # Pair1x49 labels currently cover only the priority block: first 750 numeric
93
+ # train scenes inside each K/focal group (e.g. train/f18_aperture10/scene000..).
94
+ export TRACK4D360_RECAM_SCENE_OFFSET_PER_K="${TRACK4D360_RECAM_SCENE_OFFSET_PER_K:-0}"
95
+ export TRACK4D360_RECAM_SCENE_LIMIT_PER_K="${TRACK4D360_RECAM_SCENE_LIMIT_PER_K:-750}"
96
+
97
+ # Fixed post-compaction track identity count — MUST be set (the 24h/10k-step
98
+ # backendmix run had this at 65536; this launcher previously omitted it). The
99
+ # ReCamMaster CoW tracks are dense (grid stride 2, samples=0 = uncapped), so
100
+ # without a cap N varies per sample and a heavy draw spikes the track adapter's
101
+ # proj_down activation (B,N,T,hidden) -> the observed CUDA OOM at
102
+ # conv3d_track_adapter.py proj_down (1.16 GiB alloc, step ~770). Fixing N also
103
+ # pads small draws to a constant shape, eliminating per-step fragmentation. Lower
104
+ # to 32768 for extra headroom on contended nodes.
105
+ export TRACK4D360_TRACK_IDENTITY_FIXED_N="${TRACK4D360_TRACK_IDENTITY_FIXED_N:-65536}"
106
+
107
+ export TRACK4D360_SPARSE_TRACK_PROB="${TRACK4D360_SPARSE_TRACK_PROB:-1.0}"
108
+ export TRACK4D360_SPARSE_TRACK_STRIDE="${TRACK4D360_SPARSE_TRACK_STRIDE:-2|4|8}"
109
+ export TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB="${TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB:-1.0}"
110
+ export TRACK4D360_SPARSE_TRACK_BLOCK_SIZE="${TRACK4D360_SPARSE_TRACK_BLOCK_SIZE:-64}"
111
+ export TRACK4D360_MULTI_QUERY_PROB="${TRACK4D360_MULTI_QUERY_PROB:-1.0}"
112
+ export TRACK4D360_MULTI_QUERY_CHOICES="${TRACK4D360_MULTI_QUERY_CHOICES:-1|2|3|4}"
113
+ export TRACK4D360_GEOMETRY_NOISE_PROB="${TRACK4D360_GEOMETRY_NOISE_PROB:-1.0}"
114
+ export TRACK4D360_TRACK_OFFSET_STD="${TRACK4D360_TRACK_OFFSET_STD:-0.015}"
115
+ export TRACK4D360_TRACK_TEMPORAL_JITTER_STD="${TRACK4D360_TRACK_TEMPORAL_JITTER_STD:-0.006}"
116
+ export TRACK4D360_TRACK_XY_NOISE_RATIO="${TRACK4D360_TRACK_XY_NOISE_RATIO:-0.25}"
117
+ export TRACK4D360_TRACK_FRAME_DEPTH_JITTER_RATIO="${TRACK4D360_TRACK_FRAME_DEPTH_JITTER_RATIO:-0.5}"
118
+ export TRACK4D360_DEPTH_OFFSET_STD="${TRACK4D360_DEPTH_OFFSET_STD:-0.006}"
119
+ export TRACK4D360_DEPTH_SCALE_JITTER="${TRACK4D360_DEPTH_SCALE_JITTER:-0.01}"
120
+ export TRACK4D360_SOURCE_RGB_NOISE_STD="${TRACK4D360_SOURCE_RGB_NOISE_STD:-0.01}"
121
+ export TRACK4D360_SOURCE_RGB_FRAME_JITTER_STD="${TRACK4D360_SOURCE_RGB_FRAME_JITTER_STD:-0.02}"
122
+ export TRACK4D360_DYNAMIC_MASK_AUG_PROB="${TRACK4D360_DYNAMIC_MASK_AUG_PROB:-0.5}"
123
+ export TRACK4D360_DYNAMIC_MASK_MORPH_KERNEL="${TRACK4D360_DYNAMIC_MASK_MORPH_KERNEL:-5}"
124
+ export TRACK4D360_DYNAMIC_MASK_FLIP_PROB="${TRACK4D360_DYNAMIC_MASK_FLIP_PROB:-0.002}"
125
+
126
+ export dense_warp_mode="${dense_warp_mode:-hybrid_softsplat}"
127
+ export dense_warp_motion_threshold="${dense_warp_motion_threshold:-0.005}"
128
+ export dense_conditioning_noise_max_level="${dense_conditioning_noise_max_level:-0.15}"
129
+ export source_mask_boundary_depth_dropout_prob="${source_mask_boundary_depth_dropout_prob:-0.5}"
130
+ export source_mask_boundary_kernel_size="${source_mask_boundary_kernel_size:-5}"
131
+ export freeze_track4d_condition_modules="${freeze_track4d_condition_modules:-0}"
132
+
133
+ export use_gradient_checkpointing=1
134
+ export use_gradient_checkpointing_offload="${use_gradient_checkpointing_offload:-0}"
135
+ export VXF_OFFLOAD_T5_TO_CPU="${VXF_OFFLOAD_T5_TO_CPU:-1}"
136
+ export TRACK4D360_MEM_PROBE="${TRACK4D360_MEM_PROBE:-0}"
137
+ export TRACK4D360_ADAPTER_MEM_PROFILE="${TRACK4D360_ADAPTER_MEM_PROFILE:-0}"
138
+ export PYTORCH_CUDA_ALLOC_CONF="${PYTORCH_CUDA_ALLOC_CONF:-${PYTORCH_ALLOC_CONF:-expandable_segments:True}}"
139
+ export PYTORCH_ALLOC_CONF="${PYTORCH_ALLOC_CONF:-${PYTORCH_CUDA_ALLOC_CONF}}"
140
+ export TRACK4D360_DENSE_HYBRID_CACHE_FRAMES="${TRACK4D360_DENSE_HYBRID_CACHE_FRAMES:-7}"
141
+ export TRACK4D360_DENSE_HYBRID_TGT_CHUNK="${TRACK4D360_DENSE_HYBRID_TGT_CHUNK:-2}"
142
+
143
+ export NCCL_P2P_DISABLE="${NCCL_P2P_DISABLE:-0}"
144
+ export NCCL_P2P_LEVEL="${NCCL_P2P_LEVEL:-NVL}"
145
+ export NCCL_NVLS_ENABLE=0
146
+ export NCCL_NSOCKS_PERTHREAD=8
147
+ export NCCL_SOCKET_NTHREADS=8
148
+ export NCCL_MAX_NCHANNELS="${NCCL_MAX_NCHANNELS:-32}"
149
+ export NCCL_BUFFSIZE="${NCCL_BUFFSIZE:-8388608}"
150
+ export NCCL_DEBUG="${NCCL_DEBUG:-WARN}"
151
+
152
+ export TRACK4D360_CAPTION_OUTPUT_ROOT="${TRACK4D360_CAPTION_OUTPUT_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output}"
153
+ export SYN4D_CAPTION_DIR="${SYN4D_CAPTION_DIR:-${TRACK4D360_CAPTION_OUTPUT_ROOT}/Syn4D}"
154
+
155
+ echo "[raymap-tokens 14B pair1x49 next] T2V init: ${VXF_INIT}"
156
+ echo "[raymap-tokens 14B pair1x49 next] warm start: ${WARM_START}"
157
+ echo "[raymap-tokens 14B pair1x49 next] data: ${DATA_CONFIG}"
158
+ echo "[raymap-tokens 14B pair1x49 next] ReCam source49: ${RECAMMASTER_SOURCE49_ROOT}"
159
+ echo "[raymap-tokens 14B pair1x49 next] ReCam pair1x49 labels: ${RECAMMASTER_PAIR1X49_LABEL_ROOT}"
160
+ echo "[raymap-tokens 14B pair1x49 next] ReCam CoW: ${RECAMMASTER_COW_ROOT}"
161
+ echo "[raymap-tokens 14B pair1x49 next] ReCam q1 track sampling: mode=${TRACK4D360_RECAM_TRACK_SAMPLE_MODE}, stride=${TRACK4D360_RECAM_TRACK_STRIDE}, samples=${TRACK4D360_RECAM_TRACK_SAMPLES}"
162
+ echo "[raymap-tokens 14B pair1x49 next] ReCam scene window per K: offset=${TRACK4D360_RECAM_SCENE_OFFSET_PER_K}, limit=${TRACK4D360_RECAM_SCENE_LIMIT_PER_K}"
163
+ echo "[raymap-tokens 14B pair1x49 next] q policy: ReCam q1-only; synthetic choices=${TRACK4D360_MULTI_QUERY_CHOICES}"
164
+ echo "[raymap-tokens 14B pair1x49 next] sparse synthetic tracks: stride_choices=${TRACK4D360_SPARSE_TRACK_STRIDE}, block_jitter_prob=${TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB}"
165
+ echo "[raymap-tokens 14B pair1x49 next] trainables: self_attn + raymap Ecam/camera_projector + active track/dense condition modules"
166
+ echo "[raymap-tokens 14B pair1x49 next] freeze track/dense condition modules: ${freeze_track4d_condition_modules}"
167
+ echo "[raymap-tokens 14B pair1x49 next] batch: effective=32, grad_accum=1, lr=${learning_rate}, save_steps=${save_steps}"
168
+ echo "[raymap-tokens 14B pair1x49 next] memory: grad_ckpt_offload=${use_gradient_checkpointing_offload}, hybrid_cache_frames=${TRACK4D360_DENSE_HYBRID_CACHE_FRAMES}, hybrid_tgt_chunk=${TRACK4D360_DENSE_HYBRID_TGT_CHUNK}, cuda_alloc_conf=${PYTORCH_CUDA_ALLOC_CONF}"
169
+
170
+ source "${SLURM_SUBMIT_DIR}/bash_scripts/track4d_360/ismb/sbatch_common.sh"
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Next-stage 288x512 backendmix finetune with cleaned ReCamMaster pair1x49 labels.
2
+ #
3
+ # Mix: 35% ReCamMaster pair1x49, 35% Syn4D, 5% Kubric49, 5% Kubric81,
4
+ # and 20% legacy static datasets. The static 20% keeps the previous backendmix
5
+ # relative weights: DL3DV:RE10K:MVS-Synth:TartanAir = 65000:7500:12500:25000.
6
+ # ReCamMaster uses compact square512 source49 + pair1x49 VGGT-Omega scale
7
+ # labels plus CoWTracker source tracks and is q1-only by construction.
8
+ #
9
+ # sbatch_common.sh still sanity-checks for the old MVS-Synth guard string:
10
+ # exclude_sequence_ids=['0000']. This config has no MVS-Synth entry.
11
+
12
+ num_views: 49
13
+ resolution: [512, 288]
14
+
15
+ syn4d_root: ${oc.env:SYN4D_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/Syn4D}
16
+ syn4d_caption_dir: ${oc.env:SYN4D_CAPTION_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/Syn4D}
17
+ syn4d_valid_sequences: configs/syn4d_valid_sequences.txt
18
+ dl3dv_root: ${oc.env:DL3DV_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/dl3dv_tar}
19
+ re10k_root: ${oc.env:RE10K_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/processed_re10k}
20
+ mvs_synth_root: ${oc.env:MVS_SYNTH_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/processed_mvs_synth}
21
+ tartanair_root: ${oc.env:TARTANAIR_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/processed_tartanair}
22
+ kubric49_root: ${oc.env:KUBRIC49_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/track_dataset/train_preprocessed_49frames}
23
+ kubric81_root: ${oc.env:KUBRIC81_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/track_dataset/train_preprocessed_81frames}
24
+ kubric49_caption_cache_dir: ${oc.env:KUBRIC49_CAPTION_CACHE_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/kubric_train_49}
25
+ kubric81_caption_cache_dir: ${oc.env:KUBRIC81_CAPTION_CACHE_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/kubric_train_81}
26
+ recammaster_caption_cache_dir: ${oc.env:RECAMMASTER_CAPTION_CACHE_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/recammaster}
27
+ recammaster_root: ${oc.env:RECAMMASTER_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset}
28
+ multicam_fg_mask_root: ${oc.env:MULTICAM_FG_MASK_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_SAM3_FG_81f_sources_person_human_288x512}
29
+ recammaster_source49_root: ${oc.env:RECAMMASTER_SOURCE49_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_square512_source49}
30
+ recammaster_pair1x49_label_root: ${oc.env:RECAMMASTER_PAIR1X49_LABEL_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_pair1x49_square512_scale_labels}
31
+ recammaster_cow_root: ${oc.env:RECAMMASTER_COW_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_VGGT_Omega_CoWTracker_SourceTracks_all}
32
+ recammaster_track_sample_mode: ${oc.env:TRACK4D360_RECAM_TRACK_SAMPLE_MODE,grid}
33
+ recammaster_track_stride: ${oc.env:TRACK4D360_RECAM_TRACK_STRIDE,2}
34
+ recammaster_track_samples: ${oc.env:TRACK4D360_RECAM_TRACK_SAMPLES,0}
35
+ recammaster_scene_offset_per_k: ${oc.env:TRACK4D360_RECAM_SCENE_OFFSET_PER_K,0}
36
+ recammaster_scene_limit_per_k: ${oc.env:TRACK4D360_RECAM_SCENE_LIMIT_PER_K,750}
37
+
38
+ sparse_track_prob: ${oc.env:TRACK4D360_SPARSE_TRACK_PROB,1.0}
39
+ sparse_track_stride: ${oc.env:TRACK4D360_SPARSE_TRACK_STRIDE,2|4|8}
40
+ sparse_track_query_stride_mode: ${oc.env:TRACK4D360_SPARSE_TRACK_QUERY_STRIDE_MODE,query_count_floor}
41
+ sparse_track_block_jitter_prob: ${oc.env:TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB,1.0}
42
+ sparse_track_block_size: ${oc.env:TRACK4D360_SPARSE_TRACK_BLOCK_SIZE,64}
43
+ multi_query_prob: ${oc.env:TRACK4D360_MULTI_QUERY_PROB,1.0}
44
+ multi_query_choices: ${oc.env:TRACK4D360_MULTI_QUERY_CHOICES,1|2|3|4}
45
+ geometry_noise_prob: ${oc.env:TRACK4D360_GEOMETRY_NOISE_PROB,1.0}
46
+ track_offset_std: ${oc.env:TRACK4D360_TRACK_OFFSET_STD,0.015}
47
+ track_temporal_jitter_std: ${oc.env:TRACK4D360_TRACK_TEMPORAL_JITTER_STD,0.006}
48
+ track_xy_noise_ratio: ${oc.env:TRACK4D360_TRACK_XY_NOISE_RATIO,0.25}
49
+ track_frame_depth_jitter_ratio: ${oc.env:TRACK4D360_TRACK_FRAME_DEPTH_JITTER_RATIO,0.5}
50
+ depth_offset_std: ${oc.env:TRACK4D360_DEPTH_OFFSET_STD,0.006}
51
+ depth_scale_jitter: ${oc.env:TRACK4D360_DEPTH_SCALE_JITTER,0.01}
52
+ source_rgb_noise_std: ${oc.env:TRACK4D360_SOURCE_RGB_NOISE_STD,0.01}
53
+ source_rgb_frame_jitter_std: ${oc.env:TRACK4D360_SOURCE_RGB_FRAME_JITTER_STD,0.02}
54
+ dynamic_mask_aug_prob: ${oc.env:TRACK4D360_DYNAMIC_MASK_AUG_PROB,0.5}
55
+ dynamic_mask_morph_kernel: ${oc.env:TRACK4D360_DYNAMIC_MASK_MORPH_KERNEL,5}
56
+ dynamic_mask_flip_prob: ${oc.env:TRACK4D360_DYNAMIC_MASK_FLIP_PROB,0.002}
57
+
58
+ train_datasets:
59
+ # 35% ReCamMaster CoWTracker + square512 source49 + compact pair1x49 VGGT labels.
60
+ # This dataset is q1 only: it always emits track_query_indices=[0]. The
61
+ # default track density is grid stride 2 to avoid lifting every CoW pixel
62
+ # during training; override TRACK4D360_RECAM_TRACK_SAMPLE_MODE=all for dense.
63
+ - "87500 @ ReCamMasterCowVGGT_NVS(ROOT='${recammaster_root}', vggt_root='${recammaster_source49_root}', source_vggt_root='${recammaster_source49_root}', pair1x49_label_root='${recammaster_pair1x49_label_root}', geometry_source='pair1x49_label_scale_only', cow_root='${recammaster_cow_root}', fg_mask_root='${multicam_fg_mask_root}', caption_cache_dir='${recammaster_caption_cache_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=87500, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, vggt_post_crop_mode='cow_aspect', rgb_post_crop_mode='cow_aspect', track_sample_mode='${recammaster_track_sample_mode}', track_stride=${recammaster_track_stride}, track_samples=${recammaster_track_samples}, scene_offset_per_k=${recammaster_scene_offset_per_k}, scene_limit_per_k=${recammaster_scene_limit_per_k})"
64
+ # 35% Syn4D dynamic multi-camera mesh-tracked 3D.
65
+ - "87500 @ GeometryRobustnessWrapper_NVS(Syn4D_Multi_NVS(ROOT='${syn4d_root}', caption_dir='${syn4d_caption_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=87500, track_spatial_stride=1, exclude_scenes=['flying_group'], valid_sequences_file='${syn4d_valid_sequences}', use_mp4_rgb=True, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, return_dynamic_masks=True, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}'), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
66
+ # 5% Kubric 49-frame dynamic data.
67
+ - "12500 @ GeometryRobustnessWrapper_NVS(Kubric_NVS(ROOT='${kubric49_root}', caption_cache_dir='${kubric49_caption_cache_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=12500, track_spatial_stride=1, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, return_dynamic_masks=True, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}'), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
68
+ # 5% Kubric 81-frame dynamic data sampled to 49.
69
+ - "12500 @ GeometryRobustnessWrapper_NVS(Kubric_NVS(ROOT='${kubric81_root}', caption_cache_dir='${kubric81_caption_cache_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=12500, track_spatial_stride=1, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, return_dynamic_masks=True, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}'), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
70
+ # 11.82% DL3DV (real, DA3 depth). Static total is 20%; this keeps the old static relative weight.
71
+ - "29550 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${dl3dv_root}', dataset_type='dl3dv', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=29550, track_spatial_stride=1, min_overlap_ratio=0.15, min_shift_ratio=0.15, max_shift_ratio=0.5, shift_beta_ab=[2.0, 5.0], src_center_in_target_prob=0.5, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, use_da3_depth=True), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
72
+ # 1.38% RE10K (real, DA3 depth).
73
+ - "3450 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${re10k_root}', dataset_type='re10k', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=3450, track_spatial_stride=1, min_overlap_ratio=0.15, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
74
+ # 2.28% MVS-Synth (synthetic, GT depth; scene 0000 held out).
75
+ - "5700 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${mvs_synth_root}', dataset_type='mvs_synth', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=5700, track_spatial_stride=1, min_overlap_ratio=0.15, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, exclude_sequence_ids=['0000']), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
76
+ # 4.52% TartanAir (static forward-moving scenes).
77
+ - "11300 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${tartanair_root}', dataset_type='tartanair', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=11300, track_spatial_stride=1, min_overlap_ratio=0.15, min_shift_ratio=0.15, max_shift_ratio=0.5, shift_beta_ab=[2.0, 5.0], src_center_in_target_prob=1.0, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/torch_profiler/rank0.1780383301809004862.pt.trace.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ddbf7c30a739217186bd8eac8508c6c9bf190659e87eb62aa65e8258afa93f8c
3
+ size 4715028167
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/debug-internal.log ADDED
@@ -0,0 +1,95 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-06-02T06:50:20.584631685Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-06-02T06:50:20.58696336Z","level":"INFO","msg":"stream: starting","core version":"0.26.0"}
3
+ {"time":"2026-06-02T06:50:20.969725714Z","level":"INFO","msg":"stream: created new stream","id":"3241szoo"}
4
+ {"time":"2026-06-02T06:50:20.96994734Z","level":"INFO","msg":"handler: started"}
5
+ {"time":"2026-06-02T06:50:20.970735157Z","level":"INFO","msg":"stream: started"}
6
+ {"time":"2026-06-02T06:50:20.9707597Z","level":"INFO","msg":"writer: started","stream_id":"3241szoo"}
7
+ {"time":"2026-06-02T06:50:20.971093546Z","level":"INFO","msg":"sender: started"}
8
+ {"time":"2026-06-02T06:50:22.923197858Z","level":"INFO","msg":"filestream: sending request","total_files":1,"console_offset":0,"console_lines":1}
9
+ {"time":"2026-06-02T06:50:23.133364725Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
10
+ {"time":"2026-06-02T06:50:30.303531233Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":44}
11
+ {"time":"2026-06-02T06:50:30.305530822Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":47}
12
+ {"time":"2026-06-02T06:50:37.923464724Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":0,"events_lines":2,"console_offset":1,"console_lines":115,"uploaded_len":2}
13
+ {"time":"2026-06-02T06:50:38.126829678Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
14
+ {"time":"2026-06-02T06:50:52.923591564Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":2,"events_lines":2}
15
+ {"time":"2026-06-02T06:50:53.109219757Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
16
+ {"time":"2026-06-02T06:51:07.924029832Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":4,"events_lines":2}
17
+ {"time":"2026-06-02T06:51:08.114725101Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
18
+ {"time":"2026-06-02T06:51:22.92352313Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":6,"events_lines":2}
19
+ {"time":"2026-06-02T06:51:23.119009653Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
20
+ {"time":"2026-06-02T06:51:37.924474754Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":8,"events_lines":2}
21
+ {"time":"2026-06-02T06:51:38.141711785Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
22
+ {"time":"2026-06-02T06:51:52.923701659Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":10,"events_lines":2,"console_offset":116,"console_lines":1}
23
+ {"time":"2026-06-02T06:51:53.111102047Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
24
+ {"time":"2026-06-02T06:52:07.924313952Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":12,"events_lines":2,"console_offset":117,"console_lines":3}
25
+ {"time":"2026-06-02T06:52:08.119941045Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
26
+ {"time":"2026-06-02T06:52:22.924342752Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":0,"history_lines":1,"events_offset":14,"events_lines":2,"console_offset":115,"console_lines":1}
27
+ {"time":"2026-06-02T06:52:23.14398364Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
28
+ {"time":"2026-06-02T06:52:37.924171237Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":1,"history_lines":2,"events_offset":16,"events_lines":2,"console_offset":115,"console_lines":1}
29
+ {"time":"2026-06-02T06:52:38.127651722Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
30
+ {"time":"2026-06-02T06:52:52.923475326Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":3,"history_lines":1,"events_offset":18,"events_lines":2,"console_offset":115,"console_lines":1}
31
+ {"time":"2026-06-02T06:52:53.111875741Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
32
+ {"time":"2026-06-02T06:53:07.924090817Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":4,"history_lines":1,"events_offset":20,"events_lines":2,"console_offset":115,"console_lines":1}
33
+ {"time":"2026-06-02T06:53:08.123142555Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
34
+ {"time":"2026-06-02T06:53:22.924082144Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":5,"history_lines":1,"events_offset":22,"events_lines":2,"console_offset":115,"console_lines":1}
35
+ {"time":"2026-06-02T06:53:23.155915012Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
36
+ {"time":"2026-06-02T06:53:37.924127613Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":6,"history_lines":1,"events_offset":24,"events_lines":2,"console_offset":115,"console_lines":1}
37
+ {"time":"2026-06-02T06:53:38.132051519Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
38
+ {"time":"2026-06-02T06:53:52.924552016Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":7,"history_lines":1,"events_offset":26,"events_lines":2,"console_offset":115,"console_lines":1}
39
+ {"time":"2026-06-02T06:53:53.117190814Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
40
+ {"time":"2026-06-02T06:54:07.923825475Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":8,"history_lines":1,"events_offset":28,"events_lines":2,"console_offset":115,"console_lines":1}
41
+ {"time":"2026-06-02T06:54:08.143248291Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
42
+ {"time":"2026-06-02T06:54:22.924253036Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":30,"events_lines":2}
43
+ {"time":"2026-06-02T06:54:23.162918899Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
44
+ {"time":"2026-06-02T06:54:37.924269819Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":32,"events_lines":2}
45
+ {"time":"2026-06-02T06:54:38.178293682Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
46
+ {"time":"2026-06-02T06:54:52.92343901Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":34,"events_lines":2}
47
+ {"time":"2026-06-02T06:54:53.126764907Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
48
+ {"time":"2026-06-02T06:55:07.924320537Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":36,"events_lines":2}
49
+ {"time":"2026-06-02T06:55:08.132013064Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
50
+ {"time":"2026-06-02T06:55:22.924083496Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":38,"events_lines":2}
51
+ {"time":"2026-06-02T06:55:23.183084706Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
52
+ {"time":"2026-06-02T06:55:37.923701069Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":40,"events_lines":2}
53
+ {"time":"2026-06-02T06:55:38.176471756Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
54
+ {"time":"2026-06-02T06:55:52.924750892Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":9,"history_lines":1,"events_offset":42,"events_lines":2,"console_offset":115,"console_lines":1}
55
+ {"time":"2026-06-02T06:55:53.115223076Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
56
+ {"time":"2026-06-02T06:56:07.923910025Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":10,"history_lines":1,"events_offset":44,"events_lines":2,"console_offset":115,"console_lines":1}
57
+ {"time":"2026-06-02T06:56:08.170806645Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
58
+ {"time":"2026-06-02T06:56:22.924071465Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":11,"history_lines":1,"events_offset":46,"events_lines":2,"console_offset":115,"console_lines":1}
59
+ {"time":"2026-06-02T06:56:23.147054865Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
60
+ {"time":"2026-06-02T06:56:37.923990093Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":12,"history_lines":2,"events_offset":48,"events_lines":2,"console_offset":115,"console_lines":1}
61
+ {"time":"2026-06-02T06:56:38.174203624Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
62
+ {"time":"2026-06-02T06:56:52.924661661Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":14,"history_lines":1,"events_offset":50,"events_lines":2,"console_offset":115,"console_lines":1}
63
+ {"time":"2026-06-02T06:56:53.139214445Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
64
+ {"time":"2026-06-02T06:57:07.924453061Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":15,"history_lines":1,"events_offset":52,"events_lines":2,"console_offset":115,"console_lines":1}
65
+ {"time":"2026-06-02T06:57:08.169695656Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
66
+ {"time":"2026-06-02T06:57:22.924163918Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":16,"history_lines":2,"events_offset":54,"events_lines":2,"console_offset":115,"console_lines":1}
67
+ {"time":"2026-06-02T06:57:23.146405546Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
68
+ {"time":"2026-06-02T06:57:37.924201145Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":18,"history_lines":1,"events_offset":56,"events_lines":2,"console_offset":115,"console_lines":1}
69
+ {"time":"2026-06-02T06:57:38.109290006Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
70
+ {"time":"2026-06-02T06:57:52.925187457Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":19,"history_lines":1,"events_offset":58,"events_lines":2,"console_offset":115,"console_lines":1}
71
+ {"time":"2026-06-02T06:57:53.15045367Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
72
+ {"time":"2026-06-02T06:58:07.924653715Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":20,"history_lines":2,"events_offset":60,"events_lines":2,"console_offset":115,"console_lines":1}
73
+ {"time":"2026-06-02T06:58:08.180545128Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
74
+ {"time":"2026-06-02T06:58:22.923699153Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":22,"history_lines":1,"events_offset":62,"events_lines":2,"console_offset":115,"console_lines":1}
75
+ {"time":"2026-06-02T06:58:23.120052374Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
76
+ {"time":"2026-06-02T06:58:37.923835946Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":23,"history_lines":1,"events_offset":64,"events_lines":2,"console_offset":115,"console_lines":1}
77
+ {"time":"2026-06-02T06:58:38.182881949Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
78
+ {"time":"2026-06-02T06:58:52.923692065Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":24,"history_lines":1,"events_offset":66,"events_lines":2,"console_offset":115,"console_lines":1}
79
+ {"time":"2026-06-02T06:58:53.140186235Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
80
+ {"time":"2026-06-02T06:59:07.924480574Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":25,"history_lines":2,"events_offset":68,"events_lines":2,"console_offset":115,"console_lines":1}
81
+ {"time":"2026-06-02T06:59:08.135037406Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
82
+ {"time":"2026-06-02T06:59:22.923975937Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":27,"history_lines":1,"events_offset":70,"events_lines":2,"console_offset":115,"console_lines":1}
83
+ {"time":"2026-06-02T06:59:23.116919373Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
84
+ {"time":"2026-06-02T06:59:37.924640345Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":28,"history_lines":1,"events_offset":72,"events_lines":2,"console_offset":115,"console_lines":1}
85
+ {"time":"2026-06-02T06:59:38.187698331Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
86
+ {"time":"2026-06-02T06:59:52.92428249Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":29,"history_lines":1,"events_offset":74,"events_lines":2,"console_offset":115,"console_lines":1}
87
+ {"time":"2026-06-02T06:59:53.247046902Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
88
+ {"time":"2026-06-02T06:59:57.467547914Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
89
+ {"time":"2026-06-02T06:59:57.467952574Z","level":"INFO","msg":"filestream: sending request","total_files":2,"console_offset":120,"console_lines":92,"uploaded_len":3,"complete":true,"exit_code":0}
90
+ {"time":"2026-06-02T06:59:57.66002893Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
91
+ {"time":"2026-06-02T06:59:57.660797388Z","level":"INFO","msg":"handler: operation stats","stats":{}}
92
+ {"time":"2026-06-02T06:59:57.665684932Z","level":"INFO","msg":"stream: closing"}
93
+ {"time":"2026-06-02T06:59:57.665718499Z","level":"INFO","msg":"handler: closed"}
94
+ {"time":"2026-06-02T06:59:57.666134903Z","level":"INFO","msg":"sender: closed"}
95
+ {"time":"2026-06-02T06:59:57.66616015Z","level":"INFO","msg":"stream: closed"}
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/debug.log ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-06-02 06:50:19,995 INFO MainThread:248506 [wandb_setup.py:_flush():81] Current SDK version is 0.26.0
2
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_setup.py:_flush():81] Configure stats pid to 248506
3
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_init.py:setup_run_log_directory():721] Logging user logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/logs/debug.log
5
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_init.py:setup_run_log_directory():722] Logging internal logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/logs/debug-internal.log
6
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_init.py:init():848] calling init triggers
7
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_init.py:init():853] wandb.init called with sweep_config: {}
8
+ config: {'learning_rate': 2e-05, 'weight_decay': 0.01, 'num_epochs': 200, 'save_steps': 500, 'num_workers': 7, 'prefetch_factor': 2, 'dataset_base_path': None, 'dataset_metadata_path': None, 'dataset_repeat': 1, 'dataset_num_workers': 7, 'dataloader_prefetch_factor': 2, 'data_file_keys': 'image,video', 'pointmap_stats_path': None, 'model_paths': None, 'model_id_with_origin_paths': 'Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth', 'model_variant': 'vt2v', 'extra_inputs': None, 'track_xyz_normalization': 'minmax', 'track_xyz_quantile_low': 0.02, 'track_xyz_quantile_high': 0.98, 'predict_track_residual': False, 'residual_target_xyz_source': 'clean', 'residual_zero_center_normalization': False, 'residual_linear_normalization': False, 'residual_log_scale_factor': 3.0, 'stream3r_common_scale_stat': 'mean', 'explicit_motion_loss_weight': 0.0, 'explicit_motion_loss_snr_weight': True, 'explicit_motion_loss_snr_sigmoid': True, 'explicit_motion_loss_snr_sigmoid_a': 1.0, 'explicit_motion_loss_snr_sigmoid_b': -5.0, 'explicit_motion_loss_decode_checkpoint': False, 'explicit_motion_loss_decode_chunk_size': 0, 'normalize_track_with_xyz_cond_stats': False, 'prenormalized_track_video_signed_log': False, 'use_xyz_video_condition': False, 'xyz_video_condition_scale': 1.0, 'xyz_video_condition_aug_prob': 0.0, 'xyz_video_condition_noise_std': 0.0, 'xyz_video_condition_z_noise_std': 0.0, 'xyz_video_condition_frame_scale_std': 0.0, 'xyz_video_condition_dropout_prob': 0.0, 'xyz_video_condition_mode': 'xyz', 'xyz_video_condition_cfg_dropout_prob': 0.0, 'camera_condition_cfg_dropout_prob': 0.0, 'fp8_models': None, 'offload_models': None, 'v2v_concat_mode': 'width', 'v2v_channel_patch_embed_mode': 'add', 'trainable_models': 'dit.domain_embedding,dit.control_patch_embedding', 'find_unused_parameters': False, 'task': 'sft:v2v', 'drop_text_prob': 0.1, 'drop_video_prob': 0.0, 'drop_t2v_prob': None, 'drop_i2v_prob': None, 'input_video_noise_strength': 0.0, 'input_video_noise_random_max': 500, 'output_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout', 'remove_prefix_in_ckpt': 'pipe.dit.', 'save_every_n_epochs': 2, 'max_checkpoints_to_keep': 5, 'allow_existing_checkpoints': True, 'initial_step': 2530, 'max_train_steps': 2560, 'lora_base_model': None, 'lora_target_modules': 'q,k,v,o,ffn.0,ffn.2', 'lora_rank': 32, 'lora_alpha': None, 'lora_checkpoint': None, 'preset_lora_path': None, 'preset_lora_model': None, 'use_gradient_checkpointing': True, 'use_gradient_checkpointing_offload': False, 'gradient_accumulation_steps': 1, 'use_wandb': True, 'wandb_project': 'track4d-360', 'wandb_name': 'ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp', 'wandb_entity': None, 'wandb_tags': ['ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug'], 'wandb_notes': None, 'wandb_log_gradients': False, 'wandb_log_parameters': False, 'wandb_log_freq': 100, 'wandb_watch_model': False, 'debug_latents': False, 'debug_latents_every': 500, 'debug_latents_fps': 16, 'debug_latents_quality': 5, 'debug_latents_max_frames': None, 'debug_latents_outdir': 'debug_latents', 'debug_latents_vae_tiled': True, 'debug_latents_vae_tile_size': (30, 52), 'debug_latents_vae_tile_stride': (15, 26), 'debug_missing_grads': False, 'torch_distributed_debug': None, 'use_query_frame_impulse_condition': True, 'height': 288, 'width': 512, 'max_pixels': 1048576, 'num_frames': 49, 'tokenizer_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl', 'dit_finetune_modules': 'self_attn', 'track_injection_mode': 'per_block', 'track_injection_block_mode': 'concat_project', 'train_sigma_shift': 5.0, 'dummy': False, 'steps_per_epoch': 25000, 'initialize_model_on_cpu': False, 'resume_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors', 'gs_head_init_checkpoint': None, 'joint_track_prediction': False, 'joint_track_depth_prediction': False, 'joint_depth_prediction': False, 'target_depth_log_cap': 14.0, 'target_depth_norm_type': 'log_depth', 'target_depth_max_disparity': 5.0, 'vxf_init_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors', 'channel_concat_stage': None, 'use_residual_tracks': True, 'freeze_lora_after_load': False, 'skip_v2v_required_trainables': False, 'gs_only_training': False, 'freeze_track_block_injector': False, 'freeze_track4d_condition_modules': False, 'gs_head_style': 'flashworld', 'use_gs_output': False, 'gs_zero_features': False, 'gs_feat_dim': 1024, 'gs_loss_weight': 1.0, 'gs_mse_loss_weight': 1.0, 'gs_lpips_loss_weight': 0.5, 'gs_depth_loss_weight': 0.05, 'gs_abs_depth_loss_weight': 0.0, 'gs_opacity_loss_weight': 0.1, 'gs_opacity_prune_keep_frac': 1.0, 'gs_rgb_loss_weight': None, 'gs_sh_degree': 2, 'gs_num_points_per_pixel': 2, 'no_gs_latent_detach': False, 'use_gs_feedback_training': False, 'gs_scale_cap': None, 'gs_render_patch_size': None, 'gs_legacy_union_render': False, 'gs_lyra_depth_param': 'exp', 'gs_lyra_log_depth_shift': 0.0, 'gs_lyra_dnear': 0.1, 'gs_lyra_dfar': 10.0, 'gs_replay_cache_size': 0, 'gs_dynamic_mode': 'static', 'gs_track_supervision_mode': 'pointmap', 'gs_pos_loss_weight': 1.0, 'gs_dyn_depth_loss_weight': 0.5, 'gs_opa_tv_weight': 0.01, 'gs_xyz_tv_weight': 0.01, 'gs_predict_rotation': False, 'gs_disable_delta_opacity': False, 'gs_direct_pos_loss_weight': 0.0, 'gs_track_canonical_loss_weight': 0.0, 'gs_track_delta_loss_weight': 0.0, 'gs_track_final_pos_loss_weight': 0.0, 'gs_dynamic_max_t_video': 64, 'gs_assoc_track_2d_loss_weight': 0.0, 'gs_track_accel_loss_weight': 0.0, 'gs_track_loss_quantile': 0.98, 'gs_track_canonical_detach': False, 'gs_track_joint_render': False, 'cond_drop_track_prob': 0.1, 'cond_drop_dense_prob': 0.0, 'cond_drop_geo_prob': 0.1, 'use_track_null_token_fallback': False, 'use_dense_null_token_fallback': False, 'use_plucker_camera_control': True, 'enable_v2v_plucker_camera_control': True, 'train_plucker_camera_control': False, 'use_plucker_raymap_tokens': True, 'plucker_use_ray_origin': False, 'use_perlayer_camera_control': True, 'camera_injection_mode': 'projector', 'camera_plucker_encoder_mode': 'patchify', 'warped_condition_mode': 'off', 'warped_appearance_fusion': 'concat_proj', 'warped_geom_only_dense': True, 'multi_frame_warped_appearance': False, 'multi_frame_motion_threshold': 0.01, 'fixed_step_frac': None, 'fixed_num_inference_steps': 50, 'fixed_sigma': None, 'skip_diffusion_loss': False, 'use_dense_branch': True, 'dense_in_channels': 5, 'dense_proj_dim': 32, 'dense_num_residual_blocks': 0, 'dense_alpha_track': 1.0, 'use_depth_null_token': False, 'dense_conditioning_noise_max_level': 0.15, 'source_mask_boundary_depth_dropout_prob': 0.5, 'source_mask_boundary_kernel_size': 5, 'dense_warp_mode': 'hybrid_softsplat', 'dense_warp_motion_threshold': 0.005, 'dense_hybrid_disable_dynamic_masks': False, 'use_vista4d_latent_streams': False, 'use_gen3c_warp_latent_channels': False, 'disable_track_modules': False, 'dense_only_condition_modules': False, 'torch_profile_active_steps': 5, 'torch_profile_wait_steps': 3, 'torch_profile_warmup_steps': 1, 'torch_profile_repeat': 1, 'torch_profile_dir': None, 'torch_profile_record_shapes': False, 'torch_profile_memory': False, 'torch_profile_with_stack': False, 'track_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth_noconv3dres.yaml', 'track4d360_data_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml', 'syn4d_root': '/scratch/shared/beegfs/zeren/Syn4D', 'scene': None, 'sequence_idx': None, 'track_spatial_stride': 1, '_wandb': {}}
9
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_init.py:init():896] starting backend
10
+ 2026-06-02 06:50:20,574 INFO MainThread:248506 [wandb_init.py:init():911] sending inform_init request
11
+ 2026-06-02 06:50:20,582 INFO MainThread:248506 [wandb_init.py:init():919] backend started and connected
12
+ 2026-06-02 06:50:20,584 INFO MainThread:248506 [wandb_init.py:init():989] updated telemetry
13
+ 2026-06-02 06:50:20,629 INFO MainThread:248506 [wandb_init.py:init():1013] communicating run to backend with 90.0 second timeout
14
+ 2026-06-02 06:50:21,612 INFO MainThread:248506 [wandb_init.py:init():1058] starting run threads in backend
15
+ 2026-06-02 06:50:22,909 INFO MainThread:248506 [wandb_run.py:_console_start():2542] atexit reg
16
+ 2026-06-02 06:50:22,909 INFO MainThread:248506 [wandb_run.py:_redirect():2391] redirect: wrap_raw
17
+ 2026-06-02 06:50:22,909 INFO MainThread:248506 [wandb_run.py:_redirect():2460] Wrapping output streams.
18
+ 2026-06-02 06:50:22,910 INFO MainThread:248506 [wandb_run.py:_redirect():2483] Redirects installed.
19
+ 2026-06-02 06:50:22,916 INFO MainThread:248506 [wandb_init.py:init():1098] run started, returning control to user process
20
+ 2026-06-02 06:59:56,752 INFO MainThread:248506 [wandb_run.py:_finish():2309] finishing run nirvanalan/track4d-360/3241szoo
21
+ 2026-06-02 06:59:56,772 INFO MainThread:248506 [wandb_run.py:_atexit_cleanup():2508] got exitcode: 0
22
+ 2026-06-02 06:59:56,773 INFO MainThread:248506 [wandb_run.py:_restore():2490] restore
23
+ 2026-06-02 06:59:56,773 INFO MainThread:248506 [wandb_run.py:_restore():2496] restore done
24
+ 2026-06-02 06:59:57,664 INFO MainThread:248506 [wandb_run.py:_footer_sync_info():3886] logging synced files
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/files/config.yaml ADDED
@@ -0,0 +1,700 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _wandb:
2
+ value:
3
+ cli_version: 0.26.0
4
+ e:
5
+ 4n8ybj8xil1hl68ckg9qyr4h3ok6yfl3:
6
+ args:
7
+ - --model_id_with_origin_paths
8
+ - Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth
9
+ - --tokenizer_path
10
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl
11
+ - --task
12
+ - sft:v2v
13
+ - --model_variant
14
+ - vt2v
15
+ - --v2v_concat_mode
16
+ - width
17
+ - --height
18
+ - "288"
19
+ - --width
20
+ - "512"
21
+ - --num_frames
22
+ - "49"
23
+ - --track4d360_data_config
24
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml
25
+ - --steps_per_epoch
26
+ - "25000"
27
+ - --num_epochs
28
+ - "200"
29
+ - --learning_rate
30
+ - "2e-5"
31
+ - --gradient_accumulation_steps
32
+ - "1"
33
+ - --save_steps
34
+ - "500"
35
+ - --remove_prefix_in_ckpt
36
+ - pipe.dit.
37
+ - --trainable_models
38
+ - dit.domain_embedding,dit.control_patch_embedding
39
+ - --dit_finetune_modules
40
+ - self_attn
41
+ - --track_injection_mode
42
+ - per_block
43
+ - --track_injection_block_mode
44
+ - concat_project
45
+ - --track_config
46
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth_noconv3dres.yaml
47
+ - --train_sigma_shift
48
+ - "5.0"
49
+ - --use_query_frame_impulse_condition
50
+ - --use_plucker_camera_control
51
+ - --enable_v2v_plucker_camera_control
52
+ - --use_plucker_raymap_tokens
53
+ - --use_perlayer_camera_control
54
+ - --camera_injection_mode
55
+ - projector
56
+ - --camera_plucker_encoder_mode
57
+ - patchify
58
+ - --use_dense_branch
59
+ - --dense_in_channels
60
+ - "5"
61
+ - --dense_proj_dim
62
+ - "32"
63
+ - --dense_num_residual_blocks
64
+ - "0"
65
+ - --dense_alpha_track
66
+ - "1.0"
67
+ - --dense_conditioning_noise_max_level
68
+ - "0.15"
69
+ - --source_mask_boundary_depth_dropout_prob
70
+ - "0.5"
71
+ - --source_mask_boundary_kernel_size
72
+ - "5"
73
+ - --dense_warp_mode
74
+ - hybrid_softsplat
75
+ - --dense_warp_motion_threshold
76
+ - "0.005"
77
+ - --cond_drop_geo_prob
78
+ - "0.1"
79
+ - --cond_drop_track_prob
80
+ - "0.1"
81
+ - --cond_drop_dense_prob
82
+ - "0.0"
83
+ - --initial_step
84
+ - "2530"
85
+ - --output_path
86
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout
87
+ - --dataset_num_workers
88
+ - "7"
89
+ - --dataloader_prefetch_factor
90
+ - "2"
91
+ - --drop_text_prob
92
+ - "0.1"
93
+ - --debug_latents_every
94
+ - "500"
95
+ - --allow_existing_checkpoints
96
+ - --wandb_project
97
+ - track4d-360
98
+ - --wandb_name
99
+ - ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp
100
+ - --wandb_tags
101
+ - ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug
102
+ - --use_wandb
103
+ - --vxf_init_checkpoint
104
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors
105
+ - --use_gradient_checkpointing
106
+ - --resume_checkpoint
107
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors
108
+ - --max_train_steps
109
+ - "2560"
110
+ - --torch_profile_active_steps
111
+ - "5"
112
+ - --torch_profile_wait_steps
113
+ - "3"
114
+ - --torch_profile_warmup_steps
115
+ - "1"
116
+ - --torch_profile_repeat
117
+ - "1"
118
+ - --no_torch_profile_record_shapes
119
+ - --no_torch_profile_memory
120
+ codePath: scripts/wan2.1_fun/train_track4d360.py
121
+ codePathLocal: scripts/wan2.1_fun/train_track4d360.py
122
+ cpu_count: 288
123
+ cpu_count_logical: 288
124
+ cudaVersion: "12.7"
125
+ disk:
126
+ /:
127
+ total: "255795724288"
128
+ used: "527630336"
129
+ email: yushi.lan@outlook.com
130
+ executable: /projects/public/u6ew/yslan_public/yslan_project/miniforge3/envs/th29py312/bin/python3.12
131
+ git:
132
+ commit: b6681ed2127310d14bc5b93ae5f1d20df8b9e907
133
+ remote: git@github.com:NIRVANALAN/track4d-360.git
134
+ gpu: NVIDIA GH200 120GB
135
+ gpu_count: 4
136
+ gpu_nvidia:
137
+ - architecture: Hopper
138
+ cudaCores: 16896
139
+ memoryTotal: "102625181696"
140
+ name: NVIDIA GH200 120GB
141
+ uuid: GPU-5728e766-d093-23d3-0fed-a94feb031fe3
142
+ - architecture: Hopper
143
+ cudaCores: 16896
144
+ memoryTotal: "102625181696"
145
+ name: NVIDIA GH200 120GB
146
+ uuid: GPU-098a504b-2d3b-7b84-0901-6fbce487448c
147
+ - architecture: Hopper
148
+ cudaCores: 16896
149
+ memoryTotal: "102625181696"
150
+ name: NVIDIA GH200 120GB
151
+ uuid: GPU-4752a23e-8a0b-30b8-6dca-8c3bedf9c98c
152
+ - architecture: Hopper
153
+ cudaCores: 16896
154
+ memoryTotal: "102625181696"
155
+ name: NVIDIA GH200 120GB
156
+ uuid: GPU-c54beba8-b9e0-3543-a467-81ae596f05f6
157
+ host: nid010552
158
+ memory:
159
+ total: "919613407232"
160
+ os: Linux-6.4.0-150600.23.73_15.0.14-cray_shasta_c_64k-aarch64-with-glibc2.38
161
+ program: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/scripts/wan2.1_fun/train_track4d360.py
162
+ python: CPython 3.12.13
163
+ root: /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout
164
+ slurm:
165
+ cluster_name: gracehopper
166
+ conf: /var/spool/slurmd/conf-cache/slurm.conf
167
+ cpu_bind: quiet,mask_cpu:0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
168
+ cpu_bind_list: 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
169
+ cpu_bind_type: 'mask_cpu:'
170
+ cpu_bind_verbose: quiet
171
+ cpus_on_node: "288"
172
+ distribution: cyclic
173
+ gpus: "32"
174
+ gpus_on_node: "4"
175
+ gpus_per_node: "4"
176
+ gtids: "0"
177
+ job_account: brics.u6ew
178
+ job_cpus_per_node: 288(x8)
179
+ job_end_time: "1780389885"
180
+ job_gid: "1483804847"
181
+ job_gpus: 0,1,2,3
182
+ job_id: "4982943"
183
+ job_name: 14b_pair1x49_profile_ablate_bothres0_fsdp2
184
+ job_nodelist: nid[010552-010553,010558,010561-010565]
185
+ job_num_nodes: "8"
186
+ job_partition: workq
187
+ job_qos: normal
188
+ job_start_time: "1780382685"
189
+ job_uid: "1483804847"
190
+ job_user: yushi.u6ew
191
+ jobid: "4982943"
192
+ launch_node_ipaddr: 10.242.15.10
193
+ localid: "0"
194
+ mpi_type: pmi2
195
+ nnodes: "8"
196
+ nodeid: "0"
197
+ nodelist: nid[010552-010553,010558,010561-010565]
198
+ nprocs: "8"
199
+ ntasks: "8"
200
+ ntasks_per_node: "1"
201
+ oom_kill_step: "0"
202
+ prio_process: "0"
203
+ procid: "0"
204
+ srun_comm_host: 10.242.15.10
205
+ srun_comm_port: "41219"
206
+ step_gpus: 0,1,2,3
207
+ step_id: "0"
208
+ step_launcher_port: "41219"
209
+ step_nodelist: nid[010552-010553,010558,010561-010565]
210
+ step_num_nodes: "8"
211
+ step_num_tasks: "8"
212
+ step_resv_ports: 26620-26621
213
+ step_tasks_per_node: 1(x8)
214
+ stepid: "0"
215
+ submit_dir: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360
216
+ submit_host: login40
217
+ task_pid: "248425"
218
+ tasks_per_node: 1(x8)
219
+ topology_addr: global.group7.nid010552
220
+ topology_addr_pattern: switch.switch.node
221
+ umask: "0022"
222
+ startedAt: "2026-06-02T06:50:19.986166Z"
223
+ writerId: 4n8ybj8xil1hl68ckg9qyr4h3ok6yfl3
224
+ m: []
225
+ python_version: 3.12.13
226
+ t:
227
+ "1":
228
+ - 1
229
+ - 11
230
+ - 41
231
+ - 49
232
+ - 71
233
+ - 83
234
+ - 98
235
+ - 106
236
+ "2":
237
+ - 1
238
+ - 11
239
+ - 41
240
+ - 49
241
+ - 51
242
+ - 71
243
+ - 83
244
+ - 98
245
+ - 106
246
+ "3":
247
+ - 2
248
+ - 13
249
+ - 15
250
+ - 16
251
+ - 41
252
+ - 61
253
+ "4": 3.12.13
254
+ "5": 0.26.0
255
+ "6": 4.46.2
256
+ "12": 0.26.0
257
+ "13": linux-aarch64
258
+ allow_existing_checkpoints:
259
+ value: true
260
+ camera_condition_cfg_dropout_prob:
261
+ value: 0
262
+ camera_injection_mode:
263
+ value: projector
264
+ camera_plucker_encoder_mode:
265
+ value: patchify
266
+ channel_concat_stage:
267
+ value: null
268
+ cond_drop_dense_prob:
269
+ value: 0
270
+ cond_drop_geo_prob:
271
+ value: 0.1
272
+ cond_drop_track_prob:
273
+ value: 0.1
274
+ data_file_keys:
275
+ value: image,video
276
+ dataloader_prefetch_factor:
277
+ value: 2
278
+ dataset_base_path:
279
+ value: null
280
+ dataset_metadata_path:
281
+ value: null
282
+ dataset_num_workers:
283
+ value: 7
284
+ dataset_repeat:
285
+ value: 1
286
+ debug_latents:
287
+ value: false
288
+ debug_latents_every:
289
+ value: 500
290
+ debug_latents_fps:
291
+ value: 16
292
+ debug_latents_max_frames:
293
+ value: null
294
+ debug_latents_outdir:
295
+ value: debug_latents
296
+ debug_latents_quality:
297
+ value: 5
298
+ debug_latents_vae_tile_size:
299
+ value:
300
+ - 30
301
+ - 52
302
+ debug_latents_vae_tile_stride:
303
+ value:
304
+ - 15
305
+ - 26
306
+ debug_latents_vae_tiled:
307
+ value: true
308
+ debug_missing_grads:
309
+ value: false
310
+ dense_alpha_track:
311
+ value: 1
312
+ dense_conditioning_noise_max_level:
313
+ value: 0.15
314
+ dense_hybrid_disable_dynamic_masks:
315
+ value: false
316
+ dense_in_channels:
317
+ value: 5
318
+ dense_num_residual_blocks:
319
+ value: 0
320
+ dense_only_condition_modules:
321
+ value: false
322
+ dense_proj_dim:
323
+ value: 32
324
+ dense_warp_mode:
325
+ value: hybrid_softsplat
326
+ dense_warp_motion_threshold:
327
+ value: 0.005
328
+ disable_track_modules:
329
+ value: false
330
+ dit_finetune_modules:
331
+ value: self_attn
332
+ drop_i2v_prob:
333
+ value: null
334
+ drop_t2v_prob:
335
+ value: null
336
+ drop_text_prob:
337
+ value: 0.1
338
+ drop_video_prob:
339
+ value: 0
340
+ dummy:
341
+ value: false
342
+ enable_v2v_plucker_camera_control:
343
+ value: true
344
+ explicit_motion_loss_decode_checkpoint:
345
+ value: false
346
+ explicit_motion_loss_decode_chunk_size:
347
+ value: 0
348
+ explicit_motion_loss_snr_sigmoid:
349
+ value: true
350
+ explicit_motion_loss_snr_sigmoid_a:
351
+ value: 1
352
+ explicit_motion_loss_snr_sigmoid_b:
353
+ value: -5
354
+ explicit_motion_loss_snr_weight:
355
+ value: true
356
+ explicit_motion_loss_weight:
357
+ value: 0
358
+ extra_inputs:
359
+ value: null
360
+ find_unused_parameters:
361
+ value: false
362
+ fixed_num_inference_steps:
363
+ value: 50
364
+ fixed_sigma:
365
+ value: null
366
+ fixed_step_frac:
367
+ value: null
368
+ fp8_models:
369
+ value: null
370
+ freeze_lora_after_load:
371
+ value: false
372
+ freeze_track_block_injector:
373
+ value: false
374
+ freeze_track4d_condition_modules:
375
+ value: false
376
+ gradient_accumulation_steps:
377
+ value: 1
378
+ gs_abs_depth_loss_weight:
379
+ value: 0
380
+ gs_assoc_track_2d_loss_weight:
381
+ value: 0
382
+ gs_depth_loss_weight:
383
+ value: 0.05
384
+ gs_direct_pos_loss_weight:
385
+ value: 0
386
+ gs_disable_delta_opacity:
387
+ value: false
388
+ gs_dyn_depth_loss_weight:
389
+ value: 0.5
390
+ gs_dynamic_max_t_video:
391
+ value: 64
392
+ gs_dynamic_mode:
393
+ value: static
394
+ gs_feat_dim:
395
+ value: 1024
396
+ gs_head_init_checkpoint:
397
+ value: null
398
+ gs_head_style:
399
+ value: flashworld
400
+ gs_legacy_union_render:
401
+ value: false
402
+ gs_loss_weight:
403
+ value: 1
404
+ gs_lpips_loss_weight:
405
+ value: 0.5
406
+ gs_lyra_depth_param:
407
+ value: exp
408
+ gs_lyra_dfar:
409
+ value: 10
410
+ gs_lyra_dnear:
411
+ value: 0.1
412
+ gs_lyra_log_depth_shift:
413
+ value: 0
414
+ gs_mse_loss_weight:
415
+ value: 1
416
+ gs_num_points_per_pixel:
417
+ value: 2
418
+ gs_only_training:
419
+ value: false
420
+ gs_opa_tv_weight:
421
+ value: 0.01
422
+ gs_opacity_loss_weight:
423
+ value: 0.1
424
+ gs_opacity_prune_keep_frac:
425
+ value: 1
426
+ gs_pos_loss_weight:
427
+ value: 1
428
+ gs_predict_rotation:
429
+ value: false
430
+ gs_render_patch_size:
431
+ value: null
432
+ gs_replay_cache_size:
433
+ value: 0
434
+ gs_rgb_loss_weight:
435
+ value: null
436
+ gs_scale_cap:
437
+ value: null
438
+ gs_sh_degree:
439
+ value: 2
440
+ gs_track_accel_loss_weight:
441
+ value: 0
442
+ gs_track_canonical_detach:
443
+ value: false
444
+ gs_track_canonical_loss_weight:
445
+ value: 0
446
+ gs_track_delta_loss_weight:
447
+ value: 0
448
+ gs_track_final_pos_loss_weight:
449
+ value: 0
450
+ gs_track_joint_render:
451
+ value: false
452
+ gs_track_loss_quantile:
453
+ value: 0.98
454
+ gs_track_supervision_mode:
455
+ value: pointmap
456
+ gs_xyz_tv_weight:
457
+ value: 0.01
458
+ gs_zero_features:
459
+ value: false
460
+ height:
461
+ value: 288
462
+ initial_step:
463
+ value: 2530
464
+ initialize_model_on_cpu:
465
+ value: false
466
+ input_video_noise_random_max:
467
+ value: 500
468
+ input_video_noise_strength:
469
+ value: 0
470
+ joint_depth_prediction:
471
+ value: false
472
+ joint_track_depth_prediction:
473
+ value: false
474
+ joint_track_prediction:
475
+ value: false
476
+ learning_rate:
477
+ value: 2e-05
478
+ lora_alpha:
479
+ value: null
480
+ lora_base_model:
481
+ value: null
482
+ lora_checkpoint:
483
+ value: null
484
+ lora_rank:
485
+ value: 32
486
+ lora_target_modules:
487
+ value: q,k,v,o,ffn.0,ffn.2
488
+ max_checkpoints_to_keep:
489
+ value: 5
490
+ max_pixels:
491
+ value: 1048576
492
+ max_train_steps:
493
+ value: 2560
494
+ model_id_with_origin_paths:
495
+ value: Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth
496
+ model_paths:
497
+ value: null
498
+ model_variant:
499
+ value: vt2v
500
+ multi_frame_motion_threshold:
501
+ value: 0.01
502
+ multi_frame_warped_appearance:
503
+ value: false
504
+ no_gs_latent_detach:
505
+ value: false
506
+ normalize_track_with_xyz_cond_stats:
507
+ value: false
508
+ num_epochs:
509
+ value: 200
510
+ num_frames:
511
+ value: 49
512
+ num_workers:
513
+ value: 7
514
+ offload_models:
515
+ value: null
516
+ output_path:
517
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout
518
+ plucker_use_ray_origin:
519
+ value: false
520
+ pointmap_stats_path:
521
+ value: null
522
+ predict_track_residual:
523
+ value: false
524
+ prefetch_factor:
525
+ value: 2
526
+ prenormalized_track_video_signed_log:
527
+ value: false
528
+ preset_lora_model:
529
+ value: null
530
+ preset_lora_path:
531
+ value: null
532
+ remove_prefix_in_ckpt:
533
+ value: pipe.dit.
534
+ residual_linear_normalization:
535
+ value: false
536
+ residual_log_scale_factor:
537
+ value: 3
538
+ residual_target_xyz_source:
539
+ value: clean
540
+ residual_zero_center_normalization:
541
+ value: false
542
+ resume_checkpoint:
543
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors
544
+ save_every_n_epochs:
545
+ value: 2
546
+ save_steps:
547
+ value: 500
548
+ scene:
549
+ value: null
550
+ sequence_idx:
551
+ value: null
552
+ skip_diffusion_loss:
553
+ value: false
554
+ skip_v2v_required_trainables:
555
+ value: false
556
+ source_mask_boundary_depth_dropout_prob:
557
+ value: 0.5
558
+ source_mask_boundary_kernel_size:
559
+ value: 5
560
+ steps_per_epoch:
561
+ value: 25000
562
+ stream3r_common_scale_stat:
563
+ value: mean
564
+ syn4d_root:
565
+ value: /scratch/shared/beegfs/zeren/Syn4D
566
+ target_depth_log_cap:
567
+ value: 14
568
+ target_depth_max_disparity:
569
+ value: 5
570
+ target_depth_norm_type:
571
+ value: log_depth
572
+ task:
573
+ value: sft:v2v
574
+ tokenizer_path:
575
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl
576
+ torch_distributed_debug:
577
+ value: null
578
+ torch_profile_active_steps:
579
+ value: 5
580
+ torch_profile_dir:
581
+ value: null
582
+ torch_profile_memory:
583
+ value: false
584
+ torch_profile_record_shapes:
585
+ value: false
586
+ torch_profile_repeat:
587
+ value: 1
588
+ torch_profile_wait_steps:
589
+ value: 3
590
+ torch_profile_warmup_steps:
591
+ value: 1
592
+ torch_profile_with_stack:
593
+ value: false
594
+ track_config:
595
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth_noconv3dres.yaml
596
+ track_injection_block_mode:
597
+ value: concat_project
598
+ track_injection_mode:
599
+ value: per_block
600
+ track_spatial_stride:
601
+ value: 1
602
+ track_xyz_normalization:
603
+ value: minmax
604
+ track_xyz_quantile_high:
605
+ value: 0.98
606
+ track_xyz_quantile_low:
607
+ value: 0.02
608
+ track4d360_data_config:
609
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml
610
+ train_plucker_camera_control:
611
+ value: false
612
+ train_sigma_shift:
613
+ value: 5
614
+ trainable_models:
615
+ value: dit.domain_embedding,dit.control_patch_embedding
616
+ use_dense_branch:
617
+ value: true
618
+ use_dense_null_token_fallback:
619
+ value: false
620
+ use_depth_null_token:
621
+ value: false
622
+ use_gen3c_warp_latent_channels:
623
+ value: false
624
+ use_gradient_checkpointing:
625
+ value: true
626
+ use_gradient_checkpointing_offload:
627
+ value: false
628
+ use_gs_feedback_training:
629
+ value: false
630
+ use_gs_output:
631
+ value: false
632
+ use_perlayer_camera_control:
633
+ value: true
634
+ use_plucker_camera_control:
635
+ value: true
636
+ use_plucker_raymap_tokens:
637
+ value: true
638
+ use_query_frame_impulse_condition:
639
+ value: true
640
+ use_residual_tracks:
641
+ value: true
642
+ use_track_null_token_fallback:
643
+ value: false
644
+ use_vista4d_latent_streams:
645
+ value: false
646
+ use_wandb:
647
+ value: true
648
+ use_xyz_video_condition:
649
+ value: false
650
+ v2v_channel_patch_embed_mode:
651
+ value: add
652
+ v2v_concat_mode:
653
+ value: width
654
+ vxf_init_checkpoint:
655
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors
656
+ wandb_entity:
657
+ value: null
658
+ wandb_log_freq:
659
+ value: 100
660
+ wandb_log_gradients:
661
+ value: false
662
+ wandb_log_parameters:
663
+ value: false
664
+ wandb_name:
665
+ value: ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp
666
+ wandb_notes:
667
+ value: null
668
+ wandb_project:
669
+ value: track4d-360
670
+ wandb_tags:
671
+ value:
672
+ - ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug
673
+ wandb_watch_model:
674
+ value: false
675
+ warped_appearance_fusion:
676
+ value: concat_proj
677
+ warped_condition_mode:
678
+ value: "off"
679
+ warped_geom_only_dense:
680
+ value: true
681
+ weight_decay:
682
+ value: 0.01
683
+ width:
684
+ value: 512
685
+ xyz_video_condition_aug_prob:
686
+ value: 0
687
+ xyz_video_condition_cfg_dropout_prob:
688
+ value: 0
689
+ xyz_video_condition_dropout_prob:
690
+ value: 0
691
+ xyz_video_condition_frame_scale_std:
692
+ value: 0
693
+ xyz_video_condition_mode:
694
+ value: xyz
695
+ xyz_video_condition_noise_std:
696
+ value: 0
697
+ xyz_video_condition_scale:
698
+ value: 1
699
+ xyz_video_condition_z_noise_std:
700
+ value: 0
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/files/output.log ADDED
@@ -0,0 +1,212 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ✓ WandB initialized: https://wandb.ai/nirvanalan/track4d-360/runs/3241szoo
2
+ /projects/public/u6ew/yslan_public/yslan_project/miniforge3/envs/th29py312/lib/python3.12/site-packages/accelerate/utils/fsdp_utils.py:742: UserWarning: FSDP upcast of low precision parameters to fp32 (since mixed_precision != 'no') may affect the precision of model checkpoints. This effects 585 parameters: ['pipe.dit.domain_embedding', 'pipe.dit.blocks.0.self_attn.q.weight', 'pipe.dit.blocks.0.self_attn.q.bias', 'pipe.dit.blocks.0.self_attn.k.weight', 'pipe.dit.blocks.0.self_attn.k.bias', 'pipe.dit.blocks.0.self_attn.v.weight', 'pipe.dit.blocks.0.self_attn.v.bias', 'pipe.dit.blocks.0.self_attn.o.weight', 'pipe.dit.blocks.0.self_attn.o.bias', 'pipe.dit.blocks.0.self_attn.norm_q.weight', 'pipe.dit.blocks.0.self_attn.norm_k.weight', 'pipe.dit.blocks.0.camera_projector.weight', 'pipe.dit.blocks.0.camera_projector.bias', 'pipe.dit.blocks.1.self_attn.q.weight', 'pipe.dit.blocks.1.self_attn.q.bias', 'pipe.dit.blocks.1.self_attn.k.weight', 'pipe.dit.blocks.1.self_attn.k.bias', 'pipe.dit.blocks.1.self_attn.v.weight', 'pipe.dit.blocks.1.self_attn.v.bias', 'pipe.dit.blocks.1.self_attn.o.weight', 'pipe.dit.blocks.1.self_attn.o.bias', 'pipe.dit.blocks.1.self_attn.norm_q.weight', 'pipe.dit.blocks.1.self_attn.norm_k.weight', 'pipe.dit.blocks.1.camera_projector.weight', 'pipe.dit.blocks.1.camera_projector.bias', 'pipe.dit.blocks.2.self_attn.q.weight', 'pipe.dit.blocks.2.self_attn.q.bias', 'pipe.dit.blocks.2.self_attn.k.weight', 'pipe.dit.blocks.2.self_attn.k.bias', 'pipe.dit.blocks.2.self_attn.v.weight', 'pipe.dit.blocks.2.self_attn.v.bias', 'pipe.dit.blocks.2.self_attn.o.weight', 'pipe.dit.blocks.2.self_attn.o.bias', 'pipe.dit.blocks.2.self_attn.norm_q.weight', 'pipe.dit.blocks.2.self_attn.norm_k.weight', 'pipe.dit.blocks.2.camera_projector.weight', 'pipe.dit.blocks.2.camera_projector.bias', 'pipe.dit.blocks.3.self_attn.q.weight', 'pipe.dit.blocks.3.self_attn.q.bias', 'pipe.dit.blocks.3.self_attn.k.weight', 'pipe.dit.blocks.3.self_attn.k.bias', 'pipe.dit.blocks.3.self_attn.v.weight', 'pipe.dit.blocks.3.self_attn.v.bias', 'pipe.dit.blocks.3.self_attn.o.weight', 'pipe.dit.blocks.3.self_attn.o.bias', 'pipe.dit.blocks.3.self_attn.norm_q.weight', 'pipe.dit.blocks.3.self_attn.norm_k.weight', 'pipe.dit.blocks.3.camera_projector.weight', 'pipe.dit.blocks.3.camera_projector.bias', 'pipe.dit.blocks.4.self_attn.q.weight', 'pipe.dit.blocks.4.self_attn.q.bias', 'pipe.dit.blocks.4.self_attn.k.weight', 'pipe.dit.blocks.4.self_attn.k.bias', 'pipe.dit.blocks.4.self_attn.v.weight', 'pipe.dit.blocks.4.self_attn.v.bias', 'pipe.dit.blocks.4.self_attn.o.weight', 'pipe.dit.blocks.4.self_attn.o.bias', 'pipe.dit.blocks.4.self_attn.norm_q.weight', 'pipe.dit.blocks.4.self_attn.norm_k.weight', 'pipe.dit.blocks.4.camera_projector.weight', 'pipe.dit.blocks.4.camera_projector.bias', 'pipe.dit.blocks.5.self_attn.q.weight', 'pipe.dit.blocks.5.self_attn.q.bias', 'pipe.dit.blocks.5.self_attn.k.weight', 'pipe.dit.blocks.5.self_attn.k.bias', 'pipe.dit.blocks.5.self_attn.v.weight', 'pipe.dit.blocks.5.self_attn.v.bias', 'pipe.dit.blocks.5.self_attn.o.weight', 'pipe.dit.blocks.5.self_attn.o.bias', 'pipe.dit.blocks.5.self_attn.norm_q.weight', 'pipe.dit.blocks.5.self_attn.norm_k.weight', 'pipe.dit.blocks.5.camera_projector.weight', 'pipe.dit.blocks.5.camera_projector.bias', 'pipe.dit.blocks.6.self_attn.q.weight', 'pipe.dit.blocks.6.self_attn.q.bias', 'pipe.dit.blocks.6.self_attn.k.weight', 'pipe.dit.blocks.6.self_attn.k.bias', 'pipe.dit.blocks.6.self_attn.v.weight', 'pipe.dit.blocks.6.self_attn.v.bias', 'pipe.dit.blocks.6.self_attn.o.weight', 'pipe.dit.blocks.6.self_attn.o.bias', 'pipe.dit.blocks.6.self_attn.norm_q.weight', 'pipe.dit.blocks.6.self_attn.norm_k.weight', 'pipe.dit.blocks.6.camera_projector.weight', 'pipe.dit.blocks.6.camera_projector.bias', 'pipe.dit.blocks.7.self_attn.q.weight', 'pipe.dit.blocks.7.self_attn.q.bias', 'pipe.dit.blocks.7.self_attn.k.weight', 'pipe.dit.blocks.7.self_attn.k.bias', 'pipe.dit.blocks.7.self_attn.v.weight', 'pipe.dit.blocks.7.self_attn.v.bias', 'pipe.dit.blocks.7.self_attn.o.weight', 'pipe.dit.blocks.7.self_attn.o.bias', 'pipe.dit.block
3
+ warnings.warn(
4
+
5
+ ================================================================================
6
+ TRAINABLE PARAMETERS VERIFICATION
7
+ ================================================================================
8
+
9
+ Trainable modules (93 module groups):
10
+
11
+ 📦 pipe.dit.blocks.* — 40 blocks [0..39]:
12
+ Params per block: 131,107,840 × 40 = 5,244,313,600
13
+ Submodules (example from block 0):
14
+ • self_attn: 104,888,320
15
+ • camera_projector: 26,219,520
16
+
17
+ 📦 pipe.dense_target_control_encoder.conv3d_patchify:
18
+ Total params: 10,272
19
+ ├─ weight: 10,240
20
+ ├─ bias: 32
21
+
22
+ 📦 pipe.dense_target_control_encoder.null_token:
23
+ Total params: 156
24
+ pipe.dense_target_control_encoder.null_token: 156
25
+
26
+ 📦 pipe.dense_target_control_encoder.proj_up:
27
+ Total params: 5,148
28
+ ├─ weight: 4,992
29
+ ├─ bias: 156
30
+
31
+ 📦 pipe.dit._camera_patchify_conv:
32
+ Total params: 36,870
33
+ ├─ weight: 36,864
34
+ ├─ bias: 6
35
+
36
+ 📦 pipe.dit._camera_plucker_patchify_encoder:
37
+ Total params: 35,840
38
+ ├─ weight: 30,720
39
+ ├─ bias: 5,120
40
+
41
+ 📦 pipe.dit.control_patch_embedding:
42
+ Total params: 332,800
43
+ ├─ weight: 327,680
44
+ ├─ bias: 5,120
45
+
46
+ 📦 pipe.dit.domain_embedding:
47
+ Total params: 10,240
48
+ pipe.dit.domain_embedding: 10,240
49
+
50
+ 📦 pipe.track_adapter.conv3d_patchify:
51
+ Total params: 65,568
52
+ ├─ weight: 65,536
53
+ ├─ bias: 32
54
+
55
+ 📦 pipe.track_adapter.mlp:
56
+ Total params: 80,028
57
+ ├─ 0.weight: 30,888
58
+ ├─ 2.weight: 24,336
59
+ ├─ 4.weight: 24,336
60
+ ├─ 0.bias: 156
61
+ ├─ 2.bias: 156
62
+ ├─ 4.bias: 156
63
+
64
+ 📦 pipe.track_adapter.null_token:
65
+ Total params: 156
66
+ pipe.track_adapter.null_token: 156
67
+
68
+ 📦 pipe.track_adapter.proj_down:
69
+ Total params: 5,024
70
+ ├─ weight: 4,992
71
+ ├─ bias: 32
72
+
73
+ 📦 pipe.track_adapter.proj_up:
74
+ Total params: 5,148
75
+ ├─ weight: 4,992
76
+ ├─ bias: 156
77
+
78
+ 📦 pipe.track_block_injector.encoders:
79
+ Total params: 1,080,729,600
80
+ ├─ 0.weight: 27,013,120
81
+ ├─ 1.weight: 27,013,120
82
+ ├─ 10.weight: 27,013,120
83
+ ├─ 11.weight: 27,013,120
84
+ ├─ 12.weight: 27,013,120
85
+ ├─ 13.weight: 27,013,120
86
+ ├─ 14.weight: 27,013,120
87
+ ├─ 15.weight: 27,013,120
88
+ ├─ 16.weight: 27,013,120
89
+ ├─ 17.weight: 27,013,120
90
+ ├─ 18.weight: 27,013,120
91
+ ├─ 19.weight: 27,013,120
92
+ ├─ 2.weight: 27,013,120
93
+ ├─ 20.weight: 27,013,120
94
+ ├─ 21.weight: 27,013,120
95
+ ├─ 22.weight: 27,013,120
96
+ ├─ 23.weight: 27,013,120
97
+ ├─ 24.weight: 27,013,120
98
+ ├─ 25.weight: 27,013,120
99
+ ├─ 26.weight: 27,013,120
100
+ ├─ 27.weight: 27,013,120
101
+ ├─ 28.weight: 27,013,120
102
+ ├─ 29.weight: 27,013,120
103
+ ├─ 3.weight: 27,013,120
104
+ ├─ 30.weight: 27,013,120
105
+ ├─ … (55 more)
106
+
107
+ ────────────────────────────────────────────────────────────────────────────────
108
+ NOTE: FSDP is active; counts below are from the rank-local post-wrap parameter view, not a full global parameter census.
109
+ Total trainable parameters: 6,325,630,450 (6325.63M)
110
+ Total frozen parameters: 10,363,571,200 (10363.57M)
111
+ Percentage trainable (overall): 37.90%
112
+ Percentage trainable (in requested modules: dit): 33.60%
113
+ ================================================================================
114
+
115
+ [torch-profiler] rank0 enabled (wait=3, warmup=1, active=5, repeat=1) -> /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/torch_profiler
116
+ Epoch 1/200: 0%| | 29/7813 [09:09<41:00:07, 18.96s/it, opt_step=2559, loss=0.0660]
117
+ [Track4D-360] First train batch runtime flags: use_gradient_checkpointing=True, use_gradient_checkpointing_offload=False
118
+ [soft_splat_warp] WARN: tracks grid 160x294 (N=47040) != image grid 288x512 (N=147456); bilinear-resampling tracks_world to image grid. This warning fires once per process.
119
+ [track4d-runtime] rank0 patched DiTBlock.forward saw mode=track block_idx=0
120
+ [Attention] Using: Flash Attention 2 (flash_attn)
121
+
122
+ [Training] Reached max_train_steps=2560. Stopping training.
123
+
124
+ [Checkpoint] Saving 585 parameters (total 6,325,630,450 ≈ 6325.63M):
125
+
126
+ 📦 blocks.* — 40 blocks [0..39]:
127
+ Params per block: 131,107,840 × 40 = 5,244,313,600
128
+ Submodules (example from block 0):
129
+ • self_attn: 104,888,320
130
+ • camera_projector: 26,219,520
131
+
132
+ 📦 _camera_patchify_conv.bias:
133
+ Total params: 6
134
+ _camera_patchify_conv.bias: 6
135
+
136
+ 📦 _camera_patchify_conv.weight:
137
+ Total params: 36,864
138
+ _camera_patchify_conv.weight: 36,864
139
+
140
+ 📦 _camera_plucker_patchify_encoder.bias:
141
+ Total params: 5,120
142
+ _camera_plucker_patchify_encoder.bias: 5,120
143
+
144
+ 📦 _camera_plucker_patchify_encoder.weight:
145
+ Total params: 30,720
146
+ _camera_plucker_patchify_encoder.weight: 30,720
147
+
148
+ 📦 control_patch_embedding.bias:
149
+ Total params: 5,120
150
+ control_patch_embedding.bias: 5,120
151
+
152
+ 📦 control_patch_embedding.weight:
153
+ Total params: 327,680
154
+ control_patch_embedding.weight: 327,680
155
+
156
+ 📦 domain_embedding:
157
+ Total params: 10,240
158
+ domain_embedding: 10,240
159
+
160
+ 📦 pipe.dense_target_control_encoder.conv3d_patchify:
161
+ Total params: 10,272
162
+ ├─ weight: 10,240
163
+ ├─ bias: 32
164
+
165
+ 📦 pipe.dense_target_control_encoder.null_token:
166
+ Total params: 156
167
+ pipe.dense_target_control_encoder.null_token: 156
168
+
169
+ 📦 pipe.dense_target_control_encoder.proj_up:
170
+ Total params: 5,148
171
+ ├─ weight: 4,992
172
+ ├─ bias: 156
173
+
174
+ 📦 pipe.track_adapter.conv3d_patchify:
175
+ Total params: 65,568
176
+ ├─ weight: 65,536
177
+ ├─ bias: 32
178
+
179
+ 📦 pipe.track_adapter.mlp:
180
+ Total params: 80,028
181
+ ├─ 0.weight: 30,888
182
+ ├─ 2.weight: 24,336
183
+ ├─ 4.weight: 24,336
184
+ ├─ 0.bias: 156
185
+ ├─ 2.bias: 156
186
+ ├─ 4.bias: 156
187
+
188
+ 📦 pipe.track_adapter.null_token:
189
+ Total params: 156
190
+ pipe.track_adapter.null_token: 156
191
+
192
+ 📦 pipe.track_adapter.proj_down:
193
+ Total params: 5,024
194
+ ├─ weight: 4,992
195
+ ├─ bias: 32
196
+
197
+ 📦 pipe.track_adapter.proj_up:
198
+ Total params: 5,148
199
+ ├─ weight: 4,992
200
+ ├─ bias: 156
201
+
202
+ 📦 pipe.track_block_injector.encoders:
203
+ Total params: 1,080,729,600
204
+ ├─ 0.weight: 27,013,120
205
+ ├─ 1.weight: 27,013,120
206
+ ├─ 10.weight: 27,013,120
207
+ ├─ 11.weight: 27,013,120
208
+ ├─ 12.weight: 27,013,120
209
+ ├─ 13.weight: 27,013,120
210
+ ├─ 14.weight: 27,013,120
211
+ ├─ 15.weight: 27,013,120
212
+ … (72 more)
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/files/requirements.txt ADDED
@@ -0,0 +1,208 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ videox-fun==1.0.1
2
+ track4d-360==0.1.0
3
+ func_timeout==4.3.5
4
+ click==8.3.2
5
+ pip==26.0.1
6
+ nvidia-cuda-nvrtc-cu12==12.6.77
7
+ ImageIO==2.37.3
8
+ numpy==1.26.4
9
+ boto3==1.42.91
10
+ albucore==0.0.24
11
+ lpips==0.1.4
12
+ torchsde==0.2.6
13
+ tomesd==0.1.3
14
+ yarl==1.23.0
15
+ typer==0.24.1
16
+ triton==3.5.0
17
+ cycler==0.12.1
18
+ multidict==6.7.1
19
+ OpenEXR==3.4.10
20
+ soxr==1.0.0
21
+ kiwisolver==1.5.0
22
+ simsimd==6.5.16
23
+ parso==0.8.6
24
+ einops==0.8.2
25
+ ftfy==6.3.1
26
+ platformdirs==4.9.6
27
+ groovy==0.1.2
28
+ nvidia-nvtx-cu12==12.6.77
29
+ sentry-sdk==2.58.0
30
+ certifi==2026.2.25
31
+ markdown-it-py==4.0.0
32
+ uvicorn==0.38.0
33
+ nvidia-cudnn-cu12==9.10.2.21
34
+ natsort==8.4.0
35
+ awscrt==0.31.2
36
+ accelerate==1.13.0
37
+ deepspeed==0.18.9
38
+ matplotlib==3.10.8
39
+ mdurl==0.1.2
40
+ regex==2026.4.4
41
+ nvidia-cuda-cupti-cu12==12.6.80
42
+ shellingham==1.5.4
43
+ joblib==1.5.3
44
+ zipp==3.23.1
45
+ ipython_pygments_lexers==1.1.1
46
+ executing==2.2.1
47
+ safetensors==0.7.0
48
+ typing-inspection==0.4.2
49
+ httpcore==1.0.9
50
+ semantic-version==2.10.0
51
+ wandb==0.26.0
52
+ asttokens==3.0.1
53
+ datasets==4.8.4
54
+ decorator==5.2.1
55
+ attrs==26.1.0
56
+ six==1.17.0
57
+ importlib_metadata==9.0.0
58
+ py-cpuinfo==9.0.0
59
+ requests==2.33.1
60
+ xxhash==3.6.0
61
+ pytorch-lightning==2.6.1
62
+ fsspec==2026.2.0
63
+ albumentations==2.0.8
64
+ soundfile==0.13.1
65
+ torchaudio==2.9.0
66
+ torchmetrics==1.9.0
67
+ nvidia-cusolver-cu12==11.7.1.2
68
+ scipy==1.17.1
69
+ jmespath==1.1.0
70
+ threadpoolctl==3.6.0
71
+ plyfile==1.1.2
72
+ peft==0.19.1
73
+ pycparser==3.0
74
+ Markdown==3.10.2
75
+ aiosignal==1.4.0
76
+ pydantic==2.13.2
77
+ decord2==3.3.0
78
+ safehttpx==0.1.7
79
+ scikit-image==0.26.0
80
+ fonttools==4.62.1
81
+ lazy-loader==0.5
82
+ smmap==5.0.3
83
+ starlette==1.0.0
84
+ Werkzeug==3.1.8
85
+ propcache==0.4.1
86
+ tensorboard==2.20.0
87
+ rich==15.0.0
88
+ modelscope==1.36.3
89
+ absl-py==2.4.0
90
+ traitlets==5.14.3
91
+ networkx==3.6.1
92
+ hf-xet==1.4.3
93
+ h5py==3.16.0
94
+ PyYAML==6.0.3
95
+ Jinja2==3.1.6
96
+ nvidia-cusparse-cu12==12.5.4.2
97
+ idna==3.11
98
+ transformers==4.46.2
99
+ httpx==0.28.1
100
+ MarkupSafe==3.0.3
101
+ tensorboard-data-server==0.7.2
102
+ pooch==1.9.0
103
+ torch==2.9.0+cu126
104
+ soupsieve==2.8.3
105
+ numba==0.65.0
106
+ torchdiffeq==0.2.5
107
+ jedi==0.19.2
108
+ charset-normalizer==3.4.7
109
+ annotated-doc==0.0.4
110
+ gradio==6.12.0
111
+ Pygments==2.20.0
112
+ pandas==3.0.2
113
+ stack-data==0.6.3
114
+ contourpy==1.3.3
115
+ tomlkit==0.14.0
116
+ scikit-learn==1.8.0
117
+ librosa==0.11.0
118
+ nvidia-cusparselt-cu12==0.7.1
119
+ pytz==2026.1.post1
120
+ go-with-the-track==0.1.0
121
+ yunchang==0.6.4
122
+ onnxruntime==1.24.4
123
+ ninja==1.13.0
124
+ sentencepiece==0.2.1
125
+ pexpect==4.9.0
126
+ multiprocess==0.70.19
127
+ omegaconf==2.3.0
128
+ h11==0.16.0
129
+ pydub==0.25.1
130
+ nvidia-cufft-cu12==11.3.0.4
131
+ xfuser==0.4.5
132
+ tifffile==2026.3.3
133
+ msgpack==1.1.2
134
+ psutil==7.2.2
135
+ ipython==9.12.0
136
+ hf-gradio==0.4.1
137
+ diffusers==0.33.1
138
+ python-dateutil==2.9.0.post0
139
+ orjson==3.11.8
140
+ huggingface_hub==0.36.2
141
+ nvidia-cuda-runtime-cu12==12.6.77
142
+ beautifulsoup4==4.14.3
143
+ ptyprocess==0.7.0
144
+ protobuf==7.34.1
145
+ GitPython==3.1.46
146
+ sympy==1.14.0
147
+ packaging==26.1
148
+ aiohttp==3.13.5
149
+ nvidia-nvshmem-cu12==3.3.20
150
+ annotated-types==0.7.0
151
+ antlr4-python3-runtime==4.9.3
152
+ gradio_client==2.4.1
153
+ stringzilla==4.6.0
154
+ flatbuffers==25.12.19
155
+ filelock==3.28.0
156
+ imageio-ffmpeg==0.6.0
157
+ fastapi==0.136.0
158
+ nvidia-nvjitlink-cu12==12.6.85
159
+ s3transfer==0.16.0
160
+ python-multipart==0.0.26
161
+ DistVAE==0.0.0b5
162
+ urllib3==2.6.3
163
+ nvidia-nccl-cu12==2.27.5
164
+ av==15.1.0
165
+ pure_eval==0.2.3
166
+ wheel==0.46.3
167
+ dill==0.4.1
168
+ flash_attn==2.8.3
169
+ track4d-360==0.1.0
170
+ opencv-python==4.11.0.86
171
+ anyio==4.13.0
172
+ pytest==9.0.3
173
+ nvidia-cufile-cu12==1.11.1.6
174
+ lightning-utilities==0.15.3
175
+ pluggy==1.6.0
176
+ grpcio==1.80.0
177
+ cffi==2.0.0
178
+ iniconfig==2.3.0
179
+ videox-fun==1.0.1
180
+ matplotlib-inline==0.2.1
181
+ tokenizers==0.20.3
182
+ pydantic_core==2.46.2
183
+ aiohappyeyeballs==2.6.1
184
+ typing_extensions==4.15.0
185
+ timm==1.0.26
186
+ pyarrow==23.0.1
187
+ mpmath==1.3.0
188
+ prompt_toolkit==3.0.52
189
+ mediapy==1.2.6
190
+ llvmlite==0.47.0
191
+ opencv-python-headless==4.11.0.86
192
+ hjson==3.1.0
193
+ tqdm==4.67.3
194
+ torchvision==0.24.0
195
+ pyparsing==3.3.2
196
+ nvidia-cublas-cu12==12.6.4.1
197
+ botocore==1.42.91
198
+ nanobind==2.9.2
199
+ wcwidth==0.6.0
200
+ gitdb==4.0.12
201
+ frozenlist==1.8.0
202
+ brotli==1.2.0
203
+ trampoline==0.1.2
204
+ pillow==12.2.0
205
+ audioread==3.1.0
206
+ lightning==2.6.1
207
+ setuptools==82.0.1
208
+ nvidia-curand-cu12==10.3.7.77
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/files/wandb-metadata.json ADDED
@@ -0,0 +1,235 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-6.4.0-150600.23.73_15.0.14-cray_shasta_c_64k-aarch64-with-glibc2.38",
3
+ "python": "CPython 3.12.13",
4
+ "startedAt": "2026-06-02T06:50:19.986166Z",
5
+ "args": [
6
+ "--model_id_with_origin_paths",
7
+ "Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth",
8
+ "--tokenizer_path",
9
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl",
10
+ "--task",
11
+ "sft:v2v",
12
+ "--model_variant",
13
+ "vt2v",
14
+ "--v2v_concat_mode",
15
+ "width",
16
+ "--height",
17
+ "288",
18
+ "--width",
19
+ "512",
20
+ "--num_frames",
21
+ "49",
22
+ "--track4d360_data_config",
23
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml",
24
+ "--steps_per_epoch",
25
+ "25000",
26
+ "--num_epochs",
27
+ "200",
28
+ "--learning_rate",
29
+ "2e-5",
30
+ "--gradient_accumulation_steps",
31
+ "1",
32
+ "--save_steps",
33
+ "500",
34
+ "--remove_prefix_in_ckpt",
35
+ "pipe.dit.",
36
+ "--trainable_models",
37
+ "dit.domain_embedding,dit.control_patch_embedding",
38
+ "--dit_finetune_modules",
39
+ "self_attn",
40
+ "--track_injection_mode",
41
+ "per_block",
42
+ "--track_injection_block_mode",
43
+ "concat_project",
44
+ "--track_config",
45
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth_noconv3dres.yaml",
46
+ "--train_sigma_shift",
47
+ "5.0",
48
+ "--use_query_frame_impulse_condition",
49
+ "--use_plucker_camera_control",
50
+ "--enable_v2v_plucker_camera_control",
51
+ "--use_plucker_raymap_tokens",
52
+ "--use_perlayer_camera_control",
53
+ "--camera_injection_mode",
54
+ "projector",
55
+ "--camera_plucker_encoder_mode",
56
+ "patchify",
57
+ "--use_dense_branch",
58
+ "--dense_in_channels",
59
+ "5",
60
+ "--dense_proj_dim",
61
+ "32",
62
+ "--dense_num_residual_blocks",
63
+ "0",
64
+ "--dense_alpha_track",
65
+ "1.0",
66
+ "--dense_conditioning_noise_max_level",
67
+ "0.15",
68
+ "--source_mask_boundary_depth_dropout_prob",
69
+ "0.5",
70
+ "--source_mask_boundary_kernel_size",
71
+ "5",
72
+ "--dense_warp_mode",
73
+ "hybrid_softsplat",
74
+ "--dense_warp_motion_threshold",
75
+ "0.005",
76
+ "--cond_drop_geo_prob",
77
+ "0.1",
78
+ "--cond_drop_track_prob",
79
+ "0.1",
80
+ "--cond_drop_dense_prob",
81
+ "0.0",
82
+ "--initial_step",
83
+ "2530",
84
+ "--output_path",
85
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout",
86
+ "--dataset_num_workers",
87
+ "7",
88
+ "--dataloader_prefetch_factor",
89
+ "2",
90
+ "--drop_text_prob",
91
+ "0.1",
92
+ "--debug_latents_every",
93
+ "500",
94
+ "--allow_existing_checkpoints",
95
+ "--wandb_project",
96
+ "track4d-360",
97
+ "--wandb_name",
98
+ "ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp",
99
+ "--wandb_tags",
100
+ "ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug",
101
+ "--use_wandb",
102
+ "--vxf_init_checkpoint",
103
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors",
104
+ "--use_gradient_checkpointing",
105
+ "--resume_checkpoint",
106
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors",
107
+ "--max_train_steps",
108
+ "2560",
109
+ "--torch_profile_active_steps",
110
+ "5",
111
+ "--torch_profile_wait_steps",
112
+ "3",
113
+ "--torch_profile_warmup_steps",
114
+ "1",
115
+ "--torch_profile_repeat",
116
+ "1",
117
+ "--no_torch_profile_record_shapes",
118
+ "--no_torch_profile_memory"
119
+ ],
120
+ "program": "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/scripts/wan2.1_fun/train_track4d360.py",
121
+ "codePath": "scripts/wan2.1_fun/train_track4d360.py",
122
+ "codePathLocal": "scripts/wan2.1_fun/train_track4d360.py",
123
+ "git": {
124
+ "remote": "git@github.com:NIRVANALAN/track4d-360.git",
125
+ "commit": "b6681ed2127310d14bc5b93ae5f1d20df8b9e907"
126
+ },
127
+ "email": "yushi.lan@outlook.com",
128
+ "root": "/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout",
129
+ "host": "nid010552",
130
+ "executable": "/projects/public/u6ew/yslan_public/yslan_project/miniforge3/envs/th29py312/bin/python3.12",
131
+ "cpu_count": 288,
132
+ "cpu_count_logical": 288,
133
+ "gpu": "NVIDIA GH200 120GB",
134
+ "gpu_count": 4,
135
+ "disk": {
136
+ "/": {
137
+ "total": "255795724288",
138
+ "used": "527630336"
139
+ }
140
+ },
141
+ "memory": {
142
+ "total": "919613407232"
143
+ },
144
+ "gpu_nvidia": [
145
+ {
146
+ "name": "NVIDIA GH200 120GB",
147
+ "memoryTotal": "102625181696",
148
+ "cudaCores": 16896,
149
+ "architecture": "Hopper",
150
+ "uuid": "GPU-5728e766-d093-23d3-0fed-a94feb031fe3"
151
+ },
152
+ {
153
+ "name": "NVIDIA GH200 120GB",
154
+ "memoryTotal": "102625181696",
155
+ "cudaCores": 16896,
156
+ "architecture": "Hopper",
157
+ "uuid": "GPU-098a504b-2d3b-7b84-0901-6fbce487448c"
158
+ },
159
+ {
160
+ "name": "NVIDIA GH200 120GB",
161
+ "memoryTotal": "102625181696",
162
+ "cudaCores": 16896,
163
+ "architecture": "Hopper",
164
+ "uuid": "GPU-4752a23e-8a0b-30b8-6dca-8c3bedf9c98c"
165
+ },
166
+ {
167
+ "name": "NVIDIA GH200 120GB",
168
+ "memoryTotal": "102625181696",
169
+ "cudaCores": 16896,
170
+ "architecture": "Hopper",
171
+ "uuid": "GPU-c54beba8-b9e0-3543-a467-81ae596f05f6"
172
+ }
173
+ ],
174
+ "cudaVersion": "12.7",
175
+ "slurm": {
176
+ "cluster_name": "gracehopper",
177
+ "conf": "/var/spool/slurmd/conf-cache/slurm.conf",
178
+ "cpu_bind": "quiet,mask_cpu:0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF",
179
+ "cpu_bind_list": "0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF",
180
+ "cpu_bind_type": "mask_cpu:",
181
+ "cpu_bind_verbose": "quiet",
182
+ "cpus_on_node": "288",
183
+ "distribution": "cyclic",
184
+ "gpus": "32",
185
+ "gpus_on_node": "4",
186
+ "gpus_per_node": "4",
187
+ "gtids": "0",
188
+ "job_account": "brics.u6ew",
189
+ "job_cpus_per_node": "288(x8)",
190
+ "job_end_time": "1780389885",
191
+ "job_gid": "1483804847",
192
+ "job_gpus": "0,1,2,3",
193
+ "job_id": "4982943",
194
+ "job_name": "14b_pair1x49_profile_ablate_bothres0_fsdp2",
195
+ "job_nodelist": "nid[010552-010553,010558,010561-010565]",
196
+ "job_num_nodes": "8",
197
+ "job_partition": "workq",
198
+ "job_qos": "normal",
199
+ "job_start_time": "1780382685",
200
+ "job_uid": "1483804847",
201
+ "job_user": "yushi.u6ew",
202
+ "jobid": "4982943",
203
+ "launch_node_ipaddr": "10.242.15.10",
204
+ "localid": "0",
205
+ "mpi_type": "pmi2",
206
+ "nnodes": "8",
207
+ "nodeid": "0",
208
+ "nodelist": "nid[010552-010553,010558,010561-010565]",
209
+ "nprocs": "8",
210
+ "ntasks": "8",
211
+ "ntasks_per_node": "1",
212
+ "oom_kill_step": "0",
213
+ "prio_process": "0",
214
+ "procid": "0",
215
+ "srun_comm_host": "10.242.15.10",
216
+ "srun_comm_port": "41219",
217
+ "step_gpus": "0,1,2,3",
218
+ "step_id": "0",
219
+ "step_launcher_port": "41219",
220
+ "step_nodelist": "nid[010552-010553,010558,010561-010565]",
221
+ "step_num_nodes": "8",
222
+ "step_num_tasks": "8",
223
+ "step_resv_ports": "26620-26621",
224
+ "step_tasks_per_node": "1(x8)",
225
+ "stepid": "0",
226
+ "submit_dir": "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360",
227
+ "submit_host": "login40",
228
+ "task_pid": "248425",
229
+ "tasks_per_node": "1(x8)",
230
+ "topology_addr": "global.group7.nid010552",
231
+ "topology_addr_pattern": "switch.switch.node",
232
+ "umask": "0022"
233
+ },
234
+ "writerId": "4n8ybj8xil1hl68ckg9qyr4h3ok6yfl3"
235
+ }
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/files/wandb-summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"train/timing/unit/WanVideoUnit_LatentNoiseAugmentation_s":4.6193774323910475e-05,"train/hbm/post_t5_allocated_gb":28.869243621826172,"train/track/injector/block_28/grad_norm":0.003465704619884491,"train/track/injector/block_03/grad_norm":0.01044765766710043,"train/track/injector/block_34/grad_norm":0.0024946231860667467,"train/track/injector/block_30/grad_norm":0.002777846995741129,"train/track/injector/block_16/weight_norm":71.75880432128906,"train/timing/unit/WanVideoUnit_TeaCache_s":1.9860548491124064e-05,"train/track/injector/block_31/grad_norm":0.0024459627456963062,"train/track/injector/block_16/grad_norm":0.00687984237447381,"train/track/injector/block_33/weight_norm":71.65996551513672,"train/track/injector/block_37/weight_norm":71.64706420898438,"_runtime":575,"train/track/injector/block_10/grad_norm":0.008368905633687973,"train/track/injector/block_07/weight_norm":71.78174591064453,"train/timing/unit/WanVideoUnit_NoiseInitializer_s":0.00017372827278450131,"train/track/injector/block_15/weight_norm":71.72611999511719,"train/track/injector/block_20/weight_norm":71.71338653564453,"_timestamp":1.780383580238531e+09,"train/track/injector/block_22/weight_norm":71.71588897705078,"train/timing/other_s":0.8354626465588808,"train/track/injector/block_25/weight_norm":71.73038482666016,"train/track/injector/block_09/weight_norm":71.7808609008789,"train/track/injector/block_19/weight_norm":71.72639465332031,"train/track/injector/block_23/weight_norm":71.7166519165039,"train/track/injector/block_26/weight_norm":71.73136901855469,"train/t5/on_cuda":-1,"train/timing/unit/WanVideoUnit_CameraControl_NVS_s":0.0032826303504407406,"train/track/injector/block_25/grad_norm":0.0038604442961513996,"train/track/injector/block_12/grad_norm":0.0076773264445364475,"train/track/injector/block_23/grad_norm":0.004788658581674099,"train/track/injector/block_00/weight_norm":71.61990356445312,"train/track/injector/block_24/grad_norm":0.004532807506620884,"train/track/injector/block_28/weight_norm":71.74323272705078,"train/track/injector/block_35/grad_norm":0.0023304815404117107,"train/hbm/post_t5_reserved_gb":28.94489097595215,"_step":2559,"train/track/injector/block_36/grad_norm":0.002490128856152296,"train/track/injector/block_14/weight_norm":71.75389099121094,"train/track/injector/block_19/grad_norm":0.005257687531411648,"train/track/adapter/weight_norm":16.34898567199707,"train/track/injector/block_22/grad_norm":0.0061228470876812935,"train/track/injector/weight_norm_max":71.8016357421875,"train/track/injector/block_21/grad_norm":0.004948248155415058,"train/track/injector/grad_norm_mean":0.006771211978048086,"train/track/injector/block_04/weight_norm":71.78126525878906,"train/track/injector/block_07/grad_norm":0.009389299899339676,"train/timing/forward_s":5.398297315929085,"train/track/injector/block_01/grad_norm":0.01250562071800232,"train/timing/unit/WanVideoUnit_WarpedAppearance_s":5.145258910488337e-05,"train/track/injector/block_33/grad_norm":0.0026537368539720774,"train/timing/data_wait_s":0.007866659667342901,"train/track/adapter/grad_norm":0.0046793692745268345,"train/track/injector/block_06/weight_norm":71.78472137451172,"train/timing/unit/WanVideoUnit_ShapeChecker_s":5.178374703973532e-05,"train/timing/unit/WanVideoUnit_TargetVideoEncoder_s":0.8000562191009521,"train/timing/unit/WanVideoUnit_PromptEmbedder_s":0.00022612593602389097,"train/track/injector/block_31/weight_norm":71.70204162597656,"train/track/injector/block_03/weight_norm":71.8016357421875,"train/track/injector/block_11/grad_norm":0.008123037405312061,"train/track/injector/block_01/weight_norm":71.73564910888672,"train/cond_dropout/drop_dense":0.0625,"train/track/injector/block_05/weight_norm":71.79096221923828,"train/loss":0.06599140167236328,"train/timing/backward_s":5.112877536565065,"train/track/injector/block_05/grad_norm":0.009117974899709225,"train/track/injector/block_27/grad_norm":0.0034159559290856123,"train/timing/unit/WanVideoUnit_V2VConcatenate_s":0.00012627961405087262,"train/track/injector/block_37/grad_norm":0.002511002356186509,"train/timing/dit_loss_fn_s":2.68336820602417,"train/track/injector/block_24/weight_norm":71.72334289550781,"train/track/injector/block_29/grad_norm":0.0035822265781462193,"train/track/injector/block_04/grad_norm":0.009663588367402554,"train/cond_dropout/drop_tracks":0.125,"train/track/injector/block_02/weight_norm":71.77011108398438,"train/timing/unit/WanVideoUnit_ModeDropoutAugmentation_s":2.7219219191465527e-05,"train/track/injector/block_34/weight_norm":71.64427947998047,"train/track/injector/block_02/grad_norm":0.010977485217154026,"train/track/injector/block_10/weight_norm":71.777587890625,"train/track/injector/block_11/weight_norm":71.75846099853516,"train/track/injector/block_39/grad_norm":0.010027051903307438,"train/timing/unit/WanVideoUnit_TrackPointCondition_s":0.02079019695520401,"train/track/injector/block_13/weight_norm":71.73658752441406,"train/track/injector/block_38/weight_norm":71.60935974121094,"train/track/injector/block_06/grad_norm":0.00888015516102314,"train/track/injector/block_13/grad_norm":0.007709243334829807,"train/hbm/allocated_gb":3.43538236618042,"train/timing/unit/WanVideoUnit_Vista4DStreams_s":5.778144986834377e-05,"train/track/injector/block_18/grad_norm":0.005994467996060848,"train/timing/reduction_s":0.15421424619853497,"train/track/injector/grad_norm_max":0.035497479140758514,"train/track/injector/block_18/weight_norm":71.7178955078125,"train/track/injector/block_30/weight_norm":71.70831298828125,"train/track/injector/weight_norm_mean":71.72107696533203,"train/track/injector/block_27/weight_norm":71.70165252685547,"_wandb":{"runtime":575},"train/timing/unit/WanVideoUnit_DenseTargetControl_s":0.26256734132766724,"train/track/injector/block_15/grad_norm":0.006465190555900335,"train/track/injector/block_38/grad_norm":0.003349488601088524,"train/track/injector/block_29/weight_norm":71.70823669433594,"train/hbm/reserved_gb":39.54901123046875,"train/track/injector/block_32/weight_norm":71.66661071777344,"train/track/injector/block_09/grad_norm":0.008560985326766968,"train/track/injector/block_39/weight_norm":71.61332702636719,"train/track/injector/block_35/weight_norm":71.6831283569336,"train/track/injector/block_14/grad_norm":0.007657380774617195,"train/track/injector/block_00/grad_norm":0.035497479140758514,"train/track/injector/block_21/weight_norm":71.712646484375,"train/track/injector/block_08/grad_norm":0.009059657342731953,"train/track/injector/block_17/weight_norm":71.72586822509766,"train/epoch":0,"train/timing/microsteps":1,"train/timing/unit/WanVideoUnit_UnifiedSequenceParallel_s":2.6808316761162132e-05,"train/timing/unit/WanVideoUnit_CfgMerger_s":1.3150616723578423e-05,"train/track/injector/block_17/grad_norm":0.00607278011739254,"train/track/injector/block_26/grad_norm":0.0038510886952281,"train/track/injector/block_08/weight_norm":71.77731323242188,"train/track/injector/block_20/grad_norm":0.004393440671265125,"train/timing/step_total_s":11.538062416017056,"train/track/injector/block_12/weight_norm":71.75679016113281,"train/track/injector/block_36/weight_norm":71.64958190917969,"train/timing/optimizer_s":0.03721067076548934,"train/timing/unit/WanVideoUnit_InputVideoEmbedder_XYZCondition_s":1.5892174243927002,"train/track/injector/block_32/grad_norm":0.0025009731762111187}
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/logs/debug-internal.log ADDED
@@ -0,0 +1,95 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-06-02T06:50:20.584631685Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-06-02T06:50:20.58696336Z","level":"INFO","msg":"stream: starting","core version":"0.26.0"}
3
+ {"time":"2026-06-02T06:50:20.969725714Z","level":"INFO","msg":"stream: created new stream","id":"3241szoo"}
4
+ {"time":"2026-06-02T06:50:20.96994734Z","level":"INFO","msg":"handler: started"}
5
+ {"time":"2026-06-02T06:50:20.970735157Z","level":"INFO","msg":"stream: started"}
6
+ {"time":"2026-06-02T06:50:20.9707597Z","level":"INFO","msg":"writer: started","stream_id":"3241szoo"}
7
+ {"time":"2026-06-02T06:50:20.971093546Z","level":"INFO","msg":"sender: started"}
8
+ {"time":"2026-06-02T06:50:22.923197858Z","level":"INFO","msg":"filestream: sending request","total_files":1,"console_offset":0,"console_lines":1}
9
+ {"time":"2026-06-02T06:50:23.133364725Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
10
+ {"time":"2026-06-02T06:50:30.303531233Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":44}
11
+ {"time":"2026-06-02T06:50:30.305530822Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":47}
12
+ {"time":"2026-06-02T06:50:37.923464724Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":0,"events_lines":2,"console_offset":1,"console_lines":115,"uploaded_len":2}
13
+ {"time":"2026-06-02T06:50:38.126829678Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
14
+ {"time":"2026-06-02T06:50:52.923591564Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":2,"events_lines":2}
15
+ {"time":"2026-06-02T06:50:53.109219757Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
16
+ {"time":"2026-06-02T06:51:07.924029832Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":4,"events_lines":2}
17
+ {"time":"2026-06-02T06:51:08.114725101Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
18
+ {"time":"2026-06-02T06:51:22.92352313Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":6,"events_lines":2}
19
+ {"time":"2026-06-02T06:51:23.119009653Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
20
+ {"time":"2026-06-02T06:51:37.924474754Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":8,"events_lines":2}
21
+ {"time":"2026-06-02T06:51:38.141711785Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
22
+ {"time":"2026-06-02T06:51:52.923701659Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":10,"events_lines":2,"console_offset":116,"console_lines":1}
23
+ {"time":"2026-06-02T06:51:53.111102047Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
24
+ {"time":"2026-06-02T06:52:07.924313952Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":12,"events_lines":2,"console_offset":117,"console_lines":3}
25
+ {"time":"2026-06-02T06:52:08.119941045Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
26
+ {"time":"2026-06-02T06:52:22.924342752Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":0,"history_lines":1,"events_offset":14,"events_lines":2,"console_offset":115,"console_lines":1}
27
+ {"time":"2026-06-02T06:52:23.14398364Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
28
+ {"time":"2026-06-02T06:52:37.924171237Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":1,"history_lines":2,"events_offset":16,"events_lines":2,"console_offset":115,"console_lines":1}
29
+ {"time":"2026-06-02T06:52:38.127651722Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
30
+ {"time":"2026-06-02T06:52:52.923475326Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":3,"history_lines":1,"events_offset":18,"events_lines":2,"console_offset":115,"console_lines":1}
31
+ {"time":"2026-06-02T06:52:53.111875741Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
32
+ {"time":"2026-06-02T06:53:07.924090817Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":4,"history_lines":1,"events_offset":20,"events_lines":2,"console_offset":115,"console_lines":1}
33
+ {"time":"2026-06-02T06:53:08.123142555Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
34
+ {"time":"2026-06-02T06:53:22.924082144Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":5,"history_lines":1,"events_offset":22,"events_lines":2,"console_offset":115,"console_lines":1}
35
+ {"time":"2026-06-02T06:53:23.155915012Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
36
+ {"time":"2026-06-02T06:53:37.924127613Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":6,"history_lines":1,"events_offset":24,"events_lines":2,"console_offset":115,"console_lines":1}
37
+ {"time":"2026-06-02T06:53:38.132051519Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
38
+ {"time":"2026-06-02T06:53:52.924552016Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":7,"history_lines":1,"events_offset":26,"events_lines":2,"console_offset":115,"console_lines":1}
39
+ {"time":"2026-06-02T06:53:53.117190814Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
40
+ {"time":"2026-06-02T06:54:07.923825475Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":8,"history_lines":1,"events_offset":28,"events_lines":2,"console_offset":115,"console_lines":1}
41
+ {"time":"2026-06-02T06:54:08.143248291Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
42
+ {"time":"2026-06-02T06:54:22.924253036Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":30,"events_lines":2}
43
+ {"time":"2026-06-02T06:54:23.162918899Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
44
+ {"time":"2026-06-02T06:54:37.924269819Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":32,"events_lines":2}
45
+ {"time":"2026-06-02T06:54:38.178293682Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
46
+ {"time":"2026-06-02T06:54:52.92343901Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":34,"events_lines":2}
47
+ {"time":"2026-06-02T06:54:53.126764907Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
48
+ {"time":"2026-06-02T06:55:07.924320537Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":36,"events_lines":2}
49
+ {"time":"2026-06-02T06:55:08.132013064Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
50
+ {"time":"2026-06-02T06:55:22.924083496Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":38,"events_lines":2}
51
+ {"time":"2026-06-02T06:55:23.183084706Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
52
+ {"time":"2026-06-02T06:55:37.923701069Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":40,"events_lines":2}
53
+ {"time":"2026-06-02T06:55:38.176471756Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
54
+ {"time":"2026-06-02T06:55:52.924750892Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":9,"history_lines":1,"events_offset":42,"events_lines":2,"console_offset":115,"console_lines":1}
55
+ {"time":"2026-06-02T06:55:53.115223076Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
56
+ {"time":"2026-06-02T06:56:07.923910025Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":10,"history_lines":1,"events_offset":44,"events_lines":2,"console_offset":115,"console_lines":1}
57
+ {"time":"2026-06-02T06:56:08.170806645Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
58
+ {"time":"2026-06-02T06:56:22.924071465Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":11,"history_lines":1,"events_offset":46,"events_lines":2,"console_offset":115,"console_lines":1}
59
+ {"time":"2026-06-02T06:56:23.147054865Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
60
+ {"time":"2026-06-02T06:56:37.923990093Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":12,"history_lines":2,"events_offset":48,"events_lines":2,"console_offset":115,"console_lines":1}
61
+ {"time":"2026-06-02T06:56:38.174203624Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
62
+ {"time":"2026-06-02T06:56:52.924661661Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":14,"history_lines":1,"events_offset":50,"events_lines":2,"console_offset":115,"console_lines":1}
63
+ {"time":"2026-06-02T06:56:53.139214445Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
64
+ {"time":"2026-06-02T06:57:07.924453061Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":15,"history_lines":1,"events_offset":52,"events_lines":2,"console_offset":115,"console_lines":1}
65
+ {"time":"2026-06-02T06:57:08.169695656Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
66
+ {"time":"2026-06-02T06:57:22.924163918Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":16,"history_lines":2,"events_offset":54,"events_lines":2,"console_offset":115,"console_lines":1}
67
+ {"time":"2026-06-02T06:57:23.146405546Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
68
+ {"time":"2026-06-02T06:57:37.924201145Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":18,"history_lines":1,"events_offset":56,"events_lines":2,"console_offset":115,"console_lines":1}
69
+ {"time":"2026-06-02T06:57:38.109290006Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
70
+ {"time":"2026-06-02T06:57:52.925187457Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":19,"history_lines":1,"events_offset":58,"events_lines":2,"console_offset":115,"console_lines":1}
71
+ {"time":"2026-06-02T06:57:53.15045367Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
72
+ {"time":"2026-06-02T06:58:07.924653715Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":20,"history_lines":2,"events_offset":60,"events_lines":2,"console_offset":115,"console_lines":1}
73
+ {"time":"2026-06-02T06:58:08.180545128Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
74
+ {"time":"2026-06-02T06:58:22.923699153Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":22,"history_lines":1,"events_offset":62,"events_lines":2,"console_offset":115,"console_lines":1}
75
+ {"time":"2026-06-02T06:58:23.120052374Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
76
+ {"time":"2026-06-02T06:58:37.923835946Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":23,"history_lines":1,"events_offset":64,"events_lines":2,"console_offset":115,"console_lines":1}
77
+ {"time":"2026-06-02T06:58:38.182881949Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
78
+ {"time":"2026-06-02T06:58:52.923692065Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":24,"history_lines":1,"events_offset":66,"events_lines":2,"console_offset":115,"console_lines":1}
79
+ {"time":"2026-06-02T06:58:53.140186235Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
80
+ {"time":"2026-06-02T06:59:07.924480574Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":25,"history_lines":2,"events_offset":68,"events_lines":2,"console_offset":115,"console_lines":1}
81
+ {"time":"2026-06-02T06:59:08.135037406Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
82
+ {"time":"2026-06-02T06:59:22.923975937Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":27,"history_lines":1,"events_offset":70,"events_lines":2,"console_offset":115,"console_lines":1}
83
+ {"time":"2026-06-02T06:59:23.116919373Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
84
+ {"time":"2026-06-02T06:59:37.924640345Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":28,"history_lines":1,"events_offset":72,"events_lines":2,"console_offset":115,"console_lines":1}
85
+ {"time":"2026-06-02T06:59:38.187698331Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
86
+ {"time":"2026-06-02T06:59:52.92428249Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":29,"history_lines":1,"events_offset":74,"events_lines":2,"console_offset":115,"console_lines":1}
87
+ {"time":"2026-06-02T06:59:53.247046902Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
88
+ {"time":"2026-06-02T06:59:57.467547914Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
89
+ {"time":"2026-06-02T06:59:57.467952574Z","level":"INFO","msg":"filestream: sending request","total_files":2,"console_offset":120,"console_lines":92,"uploaded_len":3,"complete":true,"exit_code":0}
90
+ {"time":"2026-06-02T06:59:57.66002893Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
91
+ {"time":"2026-06-02T06:59:57.660797388Z","level":"INFO","msg":"handler: operation stats","stats":{}}
92
+ {"time":"2026-06-02T06:59:57.665684932Z","level":"INFO","msg":"stream: closing"}
93
+ {"time":"2026-06-02T06:59:57.665718499Z","level":"INFO","msg":"handler: closed"}
94
+ {"time":"2026-06-02T06:59:57.666134903Z","level":"INFO","msg":"sender: closed"}
95
+ {"time":"2026-06-02T06:59:57.66616015Z","level":"INFO","msg":"stream: closed"}
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/logs/debug.log ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-06-02 06:50:19,995 INFO MainThread:248506 [wandb_setup.py:_flush():81] Current SDK version is 0.26.0
2
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_setup.py:_flush():81] Configure stats pid to 248506
3
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_init.py:setup_run_log_directory():721] Logging user logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/logs/debug.log
5
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_init.py:setup_run_log_directory():722] Logging internal logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/logs/debug-internal.log
6
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_init.py:init():848] calling init triggers
7
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_init.py:init():853] wandb.init called with sweep_config: {}
8
+ config: {'learning_rate': 2e-05, 'weight_decay': 0.01, 'num_epochs': 200, 'save_steps': 500, 'num_workers': 7, 'prefetch_factor': 2, 'dataset_base_path': None, 'dataset_metadata_path': None, 'dataset_repeat': 1, 'dataset_num_workers': 7, 'dataloader_prefetch_factor': 2, 'data_file_keys': 'image,video', 'pointmap_stats_path': None, 'model_paths': None, 'model_id_with_origin_paths': 'Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth', 'model_variant': 'vt2v', 'extra_inputs': None, 'track_xyz_normalization': 'minmax', 'track_xyz_quantile_low': 0.02, 'track_xyz_quantile_high': 0.98, 'predict_track_residual': False, 'residual_target_xyz_source': 'clean', 'residual_zero_center_normalization': False, 'residual_linear_normalization': False, 'residual_log_scale_factor': 3.0, 'stream3r_common_scale_stat': 'mean', 'explicit_motion_loss_weight': 0.0, 'explicit_motion_loss_snr_weight': True, 'explicit_motion_loss_snr_sigmoid': True, 'explicit_motion_loss_snr_sigmoid_a': 1.0, 'explicit_motion_loss_snr_sigmoid_b': -5.0, 'explicit_motion_loss_decode_checkpoint': False, 'explicit_motion_loss_decode_chunk_size': 0, 'normalize_track_with_xyz_cond_stats': False, 'prenormalized_track_video_signed_log': False, 'use_xyz_video_condition': False, 'xyz_video_condition_scale': 1.0, 'xyz_video_condition_aug_prob': 0.0, 'xyz_video_condition_noise_std': 0.0, 'xyz_video_condition_z_noise_std': 0.0, 'xyz_video_condition_frame_scale_std': 0.0, 'xyz_video_condition_dropout_prob': 0.0, 'xyz_video_condition_mode': 'xyz', 'xyz_video_condition_cfg_dropout_prob': 0.0, 'camera_condition_cfg_dropout_prob': 0.0, 'fp8_models': None, 'offload_models': None, 'v2v_concat_mode': 'width', 'v2v_channel_patch_embed_mode': 'add', 'trainable_models': 'dit.domain_embedding,dit.control_patch_embedding', 'find_unused_parameters': False, 'task': 'sft:v2v', 'drop_text_prob': 0.1, 'drop_video_prob': 0.0, 'drop_t2v_prob': None, 'drop_i2v_prob': None, 'input_video_noise_strength': 0.0, 'input_video_noise_random_max': 500, 'output_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout', 'remove_prefix_in_ckpt': 'pipe.dit.', 'save_every_n_epochs': 2, 'max_checkpoints_to_keep': 5, 'allow_existing_checkpoints': True, 'initial_step': 2530, 'max_train_steps': 2560, 'lora_base_model': None, 'lora_target_modules': 'q,k,v,o,ffn.0,ffn.2', 'lora_rank': 32, 'lora_alpha': None, 'lora_checkpoint': None, 'preset_lora_path': None, 'preset_lora_model': None, 'use_gradient_checkpointing': True, 'use_gradient_checkpointing_offload': False, 'gradient_accumulation_steps': 1, 'use_wandb': True, 'wandb_project': 'track4d-360', 'wandb_name': 'ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp', 'wandb_entity': None, 'wandb_tags': ['ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug'], 'wandb_notes': None, 'wandb_log_gradients': False, 'wandb_log_parameters': False, 'wandb_log_freq': 100, 'wandb_watch_model': False, 'debug_latents': False, 'debug_latents_every': 500, 'debug_latents_fps': 16, 'debug_latents_quality': 5, 'debug_latents_max_frames': None, 'debug_latents_outdir': 'debug_latents', 'debug_latents_vae_tiled': True, 'debug_latents_vae_tile_size': (30, 52), 'debug_latents_vae_tile_stride': (15, 26), 'debug_missing_grads': False, 'torch_distributed_debug': None, 'use_query_frame_impulse_condition': True, 'height': 288, 'width': 512, 'max_pixels': 1048576, 'num_frames': 49, 'tokenizer_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl', 'dit_finetune_modules': 'self_attn', 'track_injection_mode': 'per_block', 'track_injection_block_mode': 'concat_project', 'train_sigma_shift': 5.0, 'dummy': False, 'steps_per_epoch': 25000, 'initialize_model_on_cpu': False, 'resume_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors', 'gs_head_init_checkpoint': None, 'joint_track_prediction': False, 'joint_track_depth_prediction': False, 'joint_depth_prediction': False, 'target_depth_log_cap': 14.0, 'target_depth_norm_type': 'log_depth', 'target_depth_max_disparity': 5.0, 'vxf_init_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors', 'channel_concat_stage': None, 'use_residual_tracks': True, 'freeze_lora_after_load': False, 'skip_v2v_required_trainables': False, 'gs_only_training': False, 'freeze_track_block_injector': False, 'freeze_track4d_condition_modules': False, 'gs_head_style': 'flashworld', 'use_gs_output': False, 'gs_zero_features': False, 'gs_feat_dim': 1024, 'gs_loss_weight': 1.0, 'gs_mse_loss_weight': 1.0, 'gs_lpips_loss_weight': 0.5, 'gs_depth_loss_weight': 0.05, 'gs_abs_depth_loss_weight': 0.0, 'gs_opacity_loss_weight': 0.1, 'gs_opacity_prune_keep_frac': 1.0, 'gs_rgb_loss_weight': None, 'gs_sh_degree': 2, 'gs_num_points_per_pixel': 2, 'no_gs_latent_detach': False, 'use_gs_feedback_training': False, 'gs_scale_cap': None, 'gs_render_patch_size': None, 'gs_legacy_union_render': False, 'gs_lyra_depth_param': 'exp', 'gs_lyra_log_depth_shift': 0.0, 'gs_lyra_dnear': 0.1, 'gs_lyra_dfar': 10.0, 'gs_replay_cache_size': 0, 'gs_dynamic_mode': 'static', 'gs_track_supervision_mode': 'pointmap', 'gs_pos_loss_weight': 1.0, 'gs_dyn_depth_loss_weight': 0.5, 'gs_opa_tv_weight': 0.01, 'gs_xyz_tv_weight': 0.01, 'gs_predict_rotation': False, 'gs_disable_delta_opacity': False, 'gs_direct_pos_loss_weight': 0.0, 'gs_track_canonical_loss_weight': 0.0, 'gs_track_delta_loss_weight': 0.0, 'gs_track_final_pos_loss_weight': 0.0, 'gs_dynamic_max_t_video': 64, 'gs_assoc_track_2d_loss_weight': 0.0, 'gs_track_accel_loss_weight': 0.0, 'gs_track_loss_quantile': 0.98, 'gs_track_canonical_detach': False, 'gs_track_joint_render': False, 'cond_drop_track_prob': 0.1, 'cond_drop_dense_prob': 0.0, 'cond_drop_geo_prob': 0.1, 'use_track_null_token_fallback': False, 'use_dense_null_token_fallback': False, 'use_plucker_camera_control': True, 'enable_v2v_plucker_camera_control': True, 'train_plucker_camera_control': False, 'use_plucker_raymap_tokens': True, 'plucker_use_ray_origin': False, 'use_perlayer_camera_control': True, 'camera_injection_mode': 'projector', 'camera_plucker_encoder_mode': 'patchify', 'warped_condition_mode': 'off', 'warped_appearance_fusion': 'concat_proj', 'warped_geom_only_dense': True, 'multi_frame_warped_appearance': False, 'multi_frame_motion_threshold': 0.01, 'fixed_step_frac': None, 'fixed_num_inference_steps': 50, 'fixed_sigma': None, 'skip_diffusion_loss': False, 'use_dense_branch': True, 'dense_in_channels': 5, 'dense_proj_dim': 32, 'dense_num_residual_blocks': 0, 'dense_alpha_track': 1.0, 'use_depth_null_token': False, 'dense_conditioning_noise_max_level': 0.15, 'source_mask_boundary_depth_dropout_prob': 0.5, 'source_mask_boundary_kernel_size': 5, 'dense_warp_mode': 'hybrid_softsplat', 'dense_warp_motion_threshold': 0.005, 'dense_hybrid_disable_dynamic_masks': False, 'use_vista4d_latent_streams': False, 'use_gen3c_warp_latent_channels': False, 'disable_track_modules': False, 'dense_only_condition_modules': False, 'torch_profile_active_steps': 5, 'torch_profile_wait_steps': 3, 'torch_profile_warmup_steps': 1, 'torch_profile_repeat': 1, 'torch_profile_dir': None, 'torch_profile_record_shapes': False, 'torch_profile_memory': False, 'torch_profile_with_stack': False, 'track_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth_noconv3dres.yaml', 'track4d360_data_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml', 'syn4d_root': '/scratch/shared/beegfs/zeren/Syn4D', 'scene': None, 'sequence_idx': None, 'track_spatial_stride': 1, '_wandb': {}}
9
+ 2026-06-02 06:50:19,996 INFO MainThread:248506 [wandb_init.py:init():896] starting backend
10
+ 2026-06-02 06:50:20,574 INFO MainThread:248506 [wandb_init.py:init():911] sending inform_init request
11
+ 2026-06-02 06:50:20,582 INFO MainThread:248506 [wandb_init.py:init():919] backend started and connected
12
+ 2026-06-02 06:50:20,584 INFO MainThread:248506 [wandb_init.py:init():989] updated telemetry
13
+ 2026-06-02 06:50:20,629 INFO MainThread:248506 [wandb_init.py:init():1013] communicating run to backend with 90.0 second timeout
14
+ 2026-06-02 06:50:21,612 INFO MainThread:248506 [wandb_init.py:init():1058] starting run threads in backend
15
+ 2026-06-02 06:50:22,909 INFO MainThread:248506 [wandb_run.py:_console_start():2542] atexit reg
16
+ 2026-06-02 06:50:22,909 INFO MainThread:248506 [wandb_run.py:_redirect():2391] redirect: wrap_raw
17
+ 2026-06-02 06:50:22,909 INFO MainThread:248506 [wandb_run.py:_redirect():2460] Wrapping output streams.
18
+ 2026-06-02 06:50:22,910 INFO MainThread:248506 [wandb_run.py:_redirect():2483] Redirects installed.
19
+ 2026-06-02 06:50:22,916 INFO MainThread:248506 [wandb_init.py:init():1098] run started, returning control to user process
20
+ 2026-06-02 06:59:56,752 INFO MainThread:248506 [wandb_run.py:_finish():2309] finishing run nirvanalan/track4d-360/3241szoo
21
+ 2026-06-02 06:59:56,772 INFO MainThread:248506 [wandb_run.py:_atexit_cleanup():2508] got exitcode: 0
22
+ 2026-06-02 06:59:56,773 INFO MainThread:248506 [wandb_run.py:_restore():2490] restore
23
+ 2026-06-02 06:59:56,773 INFO MainThread:248506 [wandb_run.py:_restore():2496] restore done
24
+ 2026-06-02 06:59:57,664 INFO MainThread:248506 [wandb_run.py:_footer_sync_info():3886] logging synced files
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_bothres0_cond_dropout/wandb/run-20260602_065019-3241szoo/run-3241szoo.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:93acd73fe6c3a00e2250c4b09d384b5416f96bc641620d896dd946bfffcd54b2
3
+ size 686914
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/ismb_sbatch_14b_8node_288x512_fsdp_raymaptokens_nextstage_recammaster_pair1x49_eb32_lr2e5.sh ADDED
@@ -0,0 +1,170 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+ #SBATCH --job-name=14b_8node_288x512_pair1x49_next_eb32_lr2e5
3
+ #SBATCH --nodes=8
4
+ #SBATCH --gpus=32
5
+ #SBATCH --gpus-per-node=4
6
+ #SBATCH --ntasks-per-node=1
7
+ #SBATCH --time=24:00:00
8
+ #SBATCH --output=out/exp_%j_%x.out
9
+
10
+ # Track4D-360 14B FSDP 288x512 next-stage finetune.
11
+ #
12
+ # Main changes from the 288p backendmix OOM-debug launcher:
13
+ # - new DATA_TAG / output directory;
14
+ # - warm-start from the latest existing 288p backendmix checkpoint;
15
+ # - remove most static datasets from DATA_CONFIG;
16
+ # - add ReCamMaster CoWTracker + square512 source49 + compact pair1x49
17
+ # VGGT-Omega labels, q1-only by dataset construction.
18
+ #
19
+ # Submit from repo root on Isambard:
20
+ # sbatch bash_scripts/track4d_360/ismb/14b/sbatch/ismb_sbatch_14b_8node_288x512_fsdp_raymaptokens_nextstage_recammaster_pair1x49_eb32_lr2e5.sh
21
+
22
+ export NNODES=8
23
+ export MODEL_SIZE=14b
24
+ export CONDA_ENV="${CONDA_ENV:-th29py312}"
25
+ export HEIGHT=288
26
+ export WIDTH=512
27
+ export NUM_FRAMES=49
28
+ export main_process_port="${main_process_port:-29595}"
29
+
30
+ export WAN_T2V14B_DIR="${WAN_T2V14B_DIR:-${SLURM_SUBMIT_DIR}/weights/Wan2.1-T2V-14B}"
31
+ export VXF_INIT="${VXF_INIT:-${WAN_T2V14B_DIR}/diffusion_pytorch_model*.safetensors}"
32
+ export MODEL_ID_PREFIX="${MODEL_ID_PREFIX:-Wan-AI/Wan2.1-T2V-14B}"
33
+ export TOKENIZER_PATH="${TOKENIZER_PATH:-${WAN_T2V14B_DIR}/google/umt5-xxl}"
34
+ export T5_PATH="${T5_PATH:-${WAN_T2V14B_DIR}/models_t5_umt5-xxl-enc-bf16.pth}"
35
+ export VAE_PATH="${VAE_PATH:-${WAN_T2V14B_DIR}/Wan2.1_VAE.pth}"
36
+
37
+ # Prefer the current 288p backendmix run as the next-stage warm start. Fall
38
+ # back to the 384p backendmix/raymap-token parents only when no 288p checkpoint exists.
39
+ export BACKENDMIX_288_PARENT_DIR="${BACKENDMIX_288_PARENT_DIR:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_warmstart384_backendmix_q1to4_s1248_jitter_norecam_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_warmstart384_backendmix_q1to4_s1248_jitter_norecam_eb32_288x512_cond_dropout}"
40
+ export BACKENDMIX_384_PARENT_DIR="${BACKENDMIX_384_PARENT_DIR:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_resume_latest_backendmix_q1to4_s1248_jitter_norecam_eb32_384x672_cond_dropout_384x672x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_resume_latest_backendmix_q1to4_s1248_jitter_norecam_eb32_384x672_cond_dropout}"
41
+ export RAYMAP_384_PARENT_DIR="${RAYMAP_384_PARENT_DIR:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_real_synth_syn4d_recam_syncam_softsplat_raymaptokens_t2v_init_scratch_eb32_cond_dropout_384x672x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_real_synth_syn4d_recam_syncam_softsplat_raymaptokens_t2v_init_scratch_eb32_cond_dropout}"
42
+ # Warm-start from the 288p backendmix run's step-9000 checkpoint — the requested
43
+ # next-stage starting point. (FSDP1/FSDP2 read it identically: warm-start loads
44
+ # the safetensors BEFORE accelerator.prepare, on the unwrapped model, so it is
45
+ # distributed-backend-independent.) Override WARM_START=... to pick another;
46
+ # falls back to latest backendmix / 384p parents if step-9000 is absent.
47
+ if [ -z "${WARM_START:-}" ]; then
48
+ if [ -f "${BACKENDMIX_288_PARENT_DIR}/step-9000.safetensors" ]; then
49
+ export WARM_START="${BACKENDMIX_288_PARENT_DIR}/step-9000.safetensors"
50
+ else
51
+ latest="$(find "${BACKENDMIX_288_PARENT_DIR}" -maxdepth 1 -name 'step-*.safetensors' 2>/dev/null | sort -V | tail -1)"
52
+ if [ -z "${latest}" ]; then
53
+ latest="$(find "${BACKENDMIX_384_PARENT_DIR}" -maxdepth 1 -name 'step-*.safetensors' 2>/dev/null | sort -V | tail -1)"
54
+ fi
55
+ if [ -z "${latest}" ]; then
56
+ latest="$(find "${RAYMAP_384_PARENT_DIR}" -maxdepth 1 -name 'step-*.safetensors' 2>/dev/null | sort -V | tail -1)"
57
+ fi
58
+ export WARM_START="${latest:-${RAYMAP_384_PARENT_DIR}/step-3500.safetensors}"
59
+ fi
60
+ fi
61
+
62
+ if ! compgen -G "${VXF_INIT}" >/dev/null; then
63
+ echo "ERROR: VXF_INIT not found: ${VXF_INIT}" >&2
64
+ exit 1
65
+ fi
66
+ [ -d "${TOKENIZER_PATH}" ] || { echo "ERROR: TOKENIZER_PATH not found: ${TOKENIZER_PATH}" >&2; exit 1; }
67
+ [ -f "${T5_PATH}" ] || { echo "ERROR: T5_PATH not found: ${T5_PATH}" >&2; exit 1; }
68
+ [ -f "${VAE_PATH}" ] || { echo "ERROR: VAE_PATH not found: ${VAE_PATH}" >&2; exit 1; }
69
+
70
+ export TARGET_EFFECTIVE_BATCH=32
71
+ export gradient_accumulation_steps=1
72
+ export learning_rate="${learning_rate:-2e-5}"
73
+ export save_steps="${save_steps:-500}"
74
+ export dataset_num_workers="${dataset_num_workers:-7}"
75
+ export TRACK4D360_STOP_AFTER_NEXT_SAVE="${TRACK4D360_STOP_AFTER_NEXT_SAVE:-0}"
76
+
77
+ export dit_finetune_modules="${dit_finetune_modules:-self_attn}"
78
+ export DIT_FINETUNE_TAG="${DIT_FINETUNE_TAG:-attn}"
79
+ export use_plucker_raymap_tokens=1
80
+ export train_plucker_camera_control=0
81
+ export PLUCKER_WANDB_TAG="${PLUCKER_WANDB_TAG:-plucker_raymap_tokens}"
82
+
83
+ export DATA_CONFIG="${DATA_CONFIG:-${SLURM_SUBMIT_DIR}/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml}"
84
+ export DATA_TAG="${DATA_TAG:-mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512}"
85
+
86
+ export RECAMMASTER_SOURCE49_ROOT="${RECAMMASTER_SOURCE49_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_square512_source49}"
87
+ export RECAMMASTER_PAIR1X49_LABEL_ROOT="${RECAMMASTER_PAIR1X49_LABEL_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_pair1x49_square512_scale_labels}"
88
+ export RECAMMASTER_COW_ROOT="${RECAMMASTER_COW_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_VGGT_Omega_CoWTracker_SourceTracks_all}"
89
+ export TRACK4D360_RECAM_TRACK_SAMPLE_MODE="${TRACK4D360_RECAM_TRACK_SAMPLE_MODE:-grid}"
90
+ export TRACK4D360_RECAM_TRACK_STRIDE="${TRACK4D360_RECAM_TRACK_STRIDE:-2}"
91
+ export TRACK4D360_RECAM_TRACK_SAMPLES="${TRACK4D360_RECAM_TRACK_SAMPLES:-0}"
92
+ # Pair1x49 labels currently cover only the priority block: first 750 numeric
93
+ # train scenes inside each K/focal group (e.g. train/f18_aperture10/scene000..).
94
+ export TRACK4D360_RECAM_SCENE_OFFSET_PER_K="${TRACK4D360_RECAM_SCENE_OFFSET_PER_K:-0}"
95
+ export TRACK4D360_RECAM_SCENE_LIMIT_PER_K="${TRACK4D360_RECAM_SCENE_LIMIT_PER_K:-750}"
96
+
97
+ # Fixed post-compaction track identity count — MUST be set (the 24h/10k-step
98
+ # backendmix run had this at 65536; this launcher previously omitted it). The
99
+ # ReCamMaster CoW tracks are dense (grid stride 2, samples=0 = uncapped), so
100
+ # without a cap N varies per sample and a heavy draw spikes the track adapter's
101
+ # proj_down activation (B,N,T,hidden) -> the observed CUDA OOM at
102
+ # conv3d_track_adapter.py proj_down (1.16 GiB alloc, step ~770). Fixing N also
103
+ # pads small draws to a constant shape, eliminating per-step fragmentation. Lower
104
+ # to 32768 for extra headroom on contended nodes.
105
+ export TRACK4D360_TRACK_IDENTITY_FIXED_N="${TRACK4D360_TRACK_IDENTITY_FIXED_N:-65536}"
106
+
107
+ export TRACK4D360_SPARSE_TRACK_PROB="${TRACK4D360_SPARSE_TRACK_PROB:-1.0}"
108
+ export TRACK4D360_SPARSE_TRACK_STRIDE="${TRACK4D360_SPARSE_TRACK_STRIDE:-2|4|8}"
109
+ export TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB="${TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB:-1.0}"
110
+ export TRACK4D360_SPARSE_TRACK_BLOCK_SIZE="${TRACK4D360_SPARSE_TRACK_BLOCK_SIZE:-64}"
111
+ export TRACK4D360_MULTI_QUERY_PROB="${TRACK4D360_MULTI_QUERY_PROB:-1.0}"
112
+ export TRACK4D360_MULTI_QUERY_CHOICES="${TRACK4D360_MULTI_QUERY_CHOICES:-1|2|3|4}"
113
+ export TRACK4D360_GEOMETRY_NOISE_PROB="${TRACK4D360_GEOMETRY_NOISE_PROB:-1.0}"
114
+ export TRACK4D360_TRACK_OFFSET_STD="${TRACK4D360_TRACK_OFFSET_STD:-0.015}"
115
+ export TRACK4D360_TRACK_TEMPORAL_JITTER_STD="${TRACK4D360_TRACK_TEMPORAL_JITTER_STD:-0.006}"
116
+ export TRACK4D360_TRACK_XY_NOISE_RATIO="${TRACK4D360_TRACK_XY_NOISE_RATIO:-0.25}"
117
+ export TRACK4D360_TRACK_FRAME_DEPTH_JITTER_RATIO="${TRACK4D360_TRACK_FRAME_DEPTH_JITTER_RATIO:-0.5}"
118
+ export TRACK4D360_DEPTH_OFFSET_STD="${TRACK4D360_DEPTH_OFFSET_STD:-0.006}"
119
+ export TRACK4D360_DEPTH_SCALE_JITTER="${TRACK4D360_DEPTH_SCALE_JITTER:-0.01}"
120
+ export TRACK4D360_SOURCE_RGB_NOISE_STD="${TRACK4D360_SOURCE_RGB_NOISE_STD:-0.01}"
121
+ export TRACK4D360_SOURCE_RGB_FRAME_JITTER_STD="${TRACK4D360_SOURCE_RGB_FRAME_JITTER_STD:-0.02}"
122
+ export TRACK4D360_DYNAMIC_MASK_AUG_PROB="${TRACK4D360_DYNAMIC_MASK_AUG_PROB:-0.5}"
123
+ export TRACK4D360_DYNAMIC_MASK_MORPH_KERNEL="${TRACK4D360_DYNAMIC_MASK_MORPH_KERNEL:-5}"
124
+ export TRACK4D360_DYNAMIC_MASK_FLIP_PROB="${TRACK4D360_DYNAMIC_MASK_FLIP_PROB:-0.002}"
125
+
126
+ export dense_warp_mode="${dense_warp_mode:-hybrid_softsplat}"
127
+ export dense_warp_motion_threshold="${dense_warp_motion_threshold:-0.005}"
128
+ export dense_conditioning_noise_max_level="${dense_conditioning_noise_max_level:-0.15}"
129
+ export source_mask_boundary_depth_dropout_prob="${source_mask_boundary_depth_dropout_prob:-0.5}"
130
+ export source_mask_boundary_kernel_size="${source_mask_boundary_kernel_size:-5}"
131
+ export freeze_track4d_condition_modules="${freeze_track4d_condition_modules:-0}"
132
+
133
+ export use_gradient_checkpointing=1
134
+ export use_gradient_checkpointing_offload="${use_gradient_checkpointing_offload:-0}"
135
+ export VXF_OFFLOAD_T5_TO_CPU="${VXF_OFFLOAD_T5_TO_CPU:-1}"
136
+ export TRACK4D360_MEM_PROBE="${TRACK4D360_MEM_PROBE:-0}"
137
+ export TRACK4D360_ADAPTER_MEM_PROFILE="${TRACK4D360_ADAPTER_MEM_PROFILE:-0}"
138
+ export PYTORCH_CUDA_ALLOC_CONF="${PYTORCH_CUDA_ALLOC_CONF:-${PYTORCH_ALLOC_CONF:-expandable_segments:True}}"
139
+ export PYTORCH_ALLOC_CONF="${PYTORCH_ALLOC_CONF:-${PYTORCH_CUDA_ALLOC_CONF}}"
140
+ export TRACK4D360_DENSE_HYBRID_CACHE_FRAMES="${TRACK4D360_DENSE_HYBRID_CACHE_FRAMES:-7}"
141
+ export TRACK4D360_DENSE_HYBRID_TGT_CHUNK="${TRACK4D360_DENSE_HYBRID_TGT_CHUNK:-2}"
142
+
143
+ export NCCL_P2P_DISABLE="${NCCL_P2P_DISABLE:-0}"
144
+ export NCCL_P2P_LEVEL="${NCCL_P2P_LEVEL:-NVL}"
145
+ export NCCL_NVLS_ENABLE=0
146
+ export NCCL_NSOCKS_PERTHREAD=8
147
+ export NCCL_SOCKET_NTHREADS=8
148
+ export NCCL_MAX_NCHANNELS="${NCCL_MAX_NCHANNELS:-32}"
149
+ export NCCL_BUFFSIZE="${NCCL_BUFFSIZE:-8388608}"
150
+ export NCCL_DEBUG="${NCCL_DEBUG:-WARN}"
151
+
152
+ export TRACK4D360_CAPTION_OUTPUT_ROOT="${TRACK4D360_CAPTION_OUTPUT_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output}"
153
+ export SYN4D_CAPTION_DIR="${SYN4D_CAPTION_DIR:-${TRACK4D360_CAPTION_OUTPUT_ROOT}/Syn4D}"
154
+
155
+ echo "[raymap-tokens 14B pair1x49 next] T2V init: ${VXF_INIT}"
156
+ echo "[raymap-tokens 14B pair1x49 next] warm start: ${WARM_START}"
157
+ echo "[raymap-tokens 14B pair1x49 next] data: ${DATA_CONFIG}"
158
+ echo "[raymap-tokens 14B pair1x49 next] ReCam source49: ${RECAMMASTER_SOURCE49_ROOT}"
159
+ echo "[raymap-tokens 14B pair1x49 next] ReCam pair1x49 labels: ${RECAMMASTER_PAIR1X49_LABEL_ROOT}"
160
+ echo "[raymap-tokens 14B pair1x49 next] ReCam CoW: ${RECAMMASTER_COW_ROOT}"
161
+ echo "[raymap-tokens 14B pair1x49 next] ReCam q1 track sampling: mode=${TRACK4D360_RECAM_TRACK_SAMPLE_MODE}, stride=${TRACK4D360_RECAM_TRACK_STRIDE}, samples=${TRACK4D360_RECAM_TRACK_SAMPLES}"
162
+ echo "[raymap-tokens 14B pair1x49 next] ReCam scene window per K: offset=${TRACK4D360_RECAM_SCENE_OFFSET_PER_K}, limit=${TRACK4D360_RECAM_SCENE_LIMIT_PER_K}"
163
+ echo "[raymap-tokens 14B pair1x49 next] q policy: ReCam q1-only; synthetic choices=${TRACK4D360_MULTI_QUERY_CHOICES}"
164
+ echo "[raymap-tokens 14B pair1x49 next] sparse synthetic tracks: stride_choices=${TRACK4D360_SPARSE_TRACK_STRIDE}, block_jitter_prob=${TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB}"
165
+ echo "[raymap-tokens 14B pair1x49 next] trainables: self_attn + raymap Ecam/camera_projector + active track/dense condition modules"
166
+ echo "[raymap-tokens 14B pair1x49 next] freeze track/dense condition modules: ${freeze_track4d_condition_modules}"
167
+ echo "[raymap-tokens 14B pair1x49 next] batch: effective=32, grad_accum=1, lr=${learning_rate}, save_steps=${save_steps}"
168
+ echo "[raymap-tokens 14B pair1x49 next] memory: grad_ckpt_offload=${use_gradient_checkpointing_offload}, hybrid_cache_frames=${TRACK4D360_DENSE_HYBRID_CACHE_FRAMES}, hybrid_tgt_chunk=${TRACK4D360_DENSE_HYBRID_TGT_CHUNK}, cuda_alloc_conf=${PYTORCH_CUDA_ALLOC_CONF}"
169
+
170
+ source "${SLURM_SUBMIT_DIR}/bash_scripts/track4d_360/ismb/sbatch_common.sh"
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Next-stage 288x512 backendmix finetune with cleaned ReCamMaster pair1x49 labels.
2
+ #
3
+ # Mix: 35% ReCamMaster pair1x49, 35% Syn4D, 5% Kubric49, 5% Kubric81,
4
+ # and 20% legacy static datasets. The static 20% keeps the previous backendmix
5
+ # relative weights: DL3DV:RE10K:MVS-Synth:TartanAir = 65000:7500:12500:25000.
6
+ # ReCamMaster uses compact square512 source49 + pair1x49 VGGT-Omega scale
7
+ # labels plus CoWTracker source tracks and is q1-only by construction.
8
+ #
9
+ # sbatch_common.sh still sanity-checks for the old MVS-Synth guard string:
10
+ # exclude_sequence_ids=['0000']. This config has no MVS-Synth entry.
11
+
12
+ num_views: 49
13
+ resolution: [512, 288]
14
+
15
+ syn4d_root: ${oc.env:SYN4D_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/Syn4D}
16
+ syn4d_caption_dir: ${oc.env:SYN4D_CAPTION_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/Syn4D}
17
+ syn4d_valid_sequences: configs/syn4d_valid_sequences.txt
18
+ dl3dv_root: ${oc.env:DL3DV_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/dl3dv_tar}
19
+ re10k_root: ${oc.env:RE10K_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/processed_re10k}
20
+ mvs_synth_root: ${oc.env:MVS_SYNTH_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/processed_mvs_synth}
21
+ tartanair_root: ${oc.env:TARTANAIR_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/processed_tartanair}
22
+ kubric49_root: ${oc.env:KUBRIC49_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/track_dataset/train_preprocessed_49frames}
23
+ kubric81_root: ${oc.env:KUBRIC81_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/track_dataset/train_preprocessed_81frames}
24
+ kubric49_caption_cache_dir: ${oc.env:KUBRIC49_CAPTION_CACHE_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/kubric_train_49}
25
+ kubric81_caption_cache_dir: ${oc.env:KUBRIC81_CAPTION_CACHE_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/kubric_train_81}
26
+ recammaster_caption_cache_dir: ${oc.env:RECAMMASTER_CAPTION_CACHE_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/recammaster}
27
+ recammaster_root: ${oc.env:RECAMMASTER_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset}
28
+ multicam_fg_mask_root: ${oc.env:MULTICAM_FG_MASK_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_SAM3_FG_81f_sources_person_human_288x512}
29
+ recammaster_source49_root: ${oc.env:RECAMMASTER_SOURCE49_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_square512_source49}
30
+ recammaster_pair1x49_label_root: ${oc.env:RECAMMASTER_PAIR1X49_LABEL_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_pair1x49_square512_scale_labels}
31
+ recammaster_cow_root: ${oc.env:RECAMMASTER_COW_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_VGGT_Omega_CoWTracker_SourceTracks_all}
32
+ recammaster_track_sample_mode: ${oc.env:TRACK4D360_RECAM_TRACK_SAMPLE_MODE,grid}
33
+ recammaster_track_stride: ${oc.env:TRACK4D360_RECAM_TRACK_STRIDE,2}
34
+ recammaster_track_samples: ${oc.env:TRACK4D360_RECAM_TRACK_SAMPLES,0}
35
+ recammaster_scene_offset_per_k: ${oc.env:TRACK4D360_RECAM_SCENE_OFFSET_PER_K,0}
36
+ recammaster_scene_limit_per_k: ${oc.env:TRACK4D360_RECAM_SCENE_LIMIT_PER_K,750}
37
+
38
+ sparse_track_prob: ${oc.env:TRACK4D360_SPARSE_TRACK_PROB,1.0}
39
+ sparse_track_stride: ${oc.env:TRACK4D360_SPARSE_TRACK_STRIDE,2|4|8}
40
+ sparse_track_query_stride_mode: ${oc.env:TRACK4D360_SPARSE_TRACK_QUERY_STRIDE_MODE,query_count_floor}
41
+ sparse_track_block_jitter_prob: ${oc.env:TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB,1.0}
42
+ sparse_track_block_size: ${oc.env:TRACK4D360_SPARSE_TRACK_BLOCK_SIZE,64}
43
+ multi_query_prob: ${oc.env:TRACK4D360_MULTI_QUERY_PROB,1.0}
44
+ multi_query_choices: ${oc.env:TRACK4D360_MULTI_QUERY_CHOICES,1|2|3|4}
45
+ geometry_noise_prob: ${oc.env:TRACK4D360_GEOMETRY_NOISE_PROB,1.0}
46
+ track_offset_std: ${oc.env:TRACK4D360_TRACK_OFFSET_STD,0.015}
47
+ track_temporal_jitter_std: ${oc.env:TRACK4D360_TRACK_TEMPORAL_JITTER_STD,0.006}
48
+ track_xy_noise_ratio: ${oc.env:TRACK4D360_TRACK_XY_NOISE_RATIO,0.25}
49
+ track_frame_depth_jitter_ratio: ${oc.env:TRACK4D360_TRACK_FRAME_DEPTH_JITTER_RATIO,0.5}
50
+ depth_offset_std: ${oc.env:TRACK4D360_DEPTH_OFFSET_STD,0.006}
51
+ depth_scale_jitter: ${oc.env:TRACK4D360_DEPTH_SCALE_JITTER,0.01}
52
+ source_rgb_noise_std: ${oc.env:TRACK4D360_SOURCE_RGB_NOISE_STD,0.01}
53
+ source_rgb_frame_jitter_std: ${oc.env:TRACK4D360_SOURCE_RGB_FRAME_JITTER_STD,0.02}
54
+ dynamic_mask_aug_prob: ${oc.env:TRACK4D360_DYNAMIC_MASK_AUG_PROB,0.5}
55
+ dynamic_mask_morph_kernel: ${oc.env:TRACK4D360_DYNAMIC_MASK_MORPH_KERNEL,5}
56
+ dynamic_mask_flip_prob: ${oc.env:TRACK4D360_DYNAMIC_MASK_FLIP_PROB,0.002}
57
+
58
+ train_datasets:
59
+ # 35% ReCamMaster CoWTracker + square512 source49 + compact pair1x49 VGGT labels.
60
+ # This dataset is q1 only: it always emits track_query_indices=[0]. The
61
+ # default track density is grid stride 2 to avoid lifting every CoW pixel
62
+ # during training; override TRACK4D360_RECAM_TRACK_SAMPLE_MODE=all for dense.
63
+ - "87500 @ ReCamMasterCowVGGT_NVS(ROOT='${recammaster_root}', vggt_root='${recammaster_source49_root}', source_vggt_root='${recammaster_source49_root}', pair1x49_label_root='${recammaster_pair1x49_label_root}', geometry_source='pair1x49_label_scale_only', cow_root='${recammaster_cow_root}', fg_mask_root='${multicam_fg_mask_root}', caption_cache_dir='${recammaster_caption_cache_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=87500, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, vggt_post_crop_mode='cow_aspect', rgb_post_crop_mode='cow_aspect', track_sample_mode='${recammaster_track_sample_mode}', track_stride=${recammaster_track_stride}, track_samples=${recammaster_track_samples}, scene_offset_per_k=${recammaster_scene_offset_per_k}, scene_limit_per_k=${recammaster_scene_limit_per_k})"
64
+ # 35% Syn4D dynamic multi-camera mesh-tracked 3D.
65
+ - "87500 @ GeometryRobustnessWrapper_NVS(Syn4D_Multi_NVS(ROOT='${syn4d_root}', caption_dir='${syn4d_caption_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=87500, track_spatial_stride=1, exclude_scenes=['flying_group'], valid_sequences_file='${syn4d_valid_sequences}', use_mp4_rgb=True, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, return_dynamic_masks=True, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}'), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
66
+ # 5% Kubric 49-frame dynamic data.
67
+ - "12500 @ GeometryRobustnessWrapper_NVS(Kubric_NVS(ROOT='${kubric49_root}', caption_cache_dir='${kubric49_caption_cache_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=12500, track_spatial_stride=1, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, return_dynamic_masks=True, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}'), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
68
+ # 5% Kubric 81-frame dynamic data sampled to 49.
69
+ - "12500 @ GeometryRobustnessWrapper_NVS(Kubric_NVS(ROOT='${kubric81_root}', caption_cache_dir='${kubric81_caption_cache_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=12500, track_spatial_stride=1, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, return_dynamic_masks=True, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}'), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
70
+ # 11.82% DL3DV (real, DA3 depth). Static total is 20%; this keeps the old static relative weight.
71
+ - "29550 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${dl3dv_root}', dataset_type='dl3dv', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=29550, track_spatial_stride=1, min_overlap_ratio=0.15, min_shift_ratio=0.15, max_shift_ratio=0.5, shift_beta_ab=[2.0, 5.0], src_center_in_target_prob=0.5, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, use_da3_depth=True), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
72
+ # 1.38% RE10K (real, DA3 depth).
73
+ - "3450 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${re10k_root}', dataset_type='re10k', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=3450, track_spatial_stride=1, min_overlap_ratio=0.15, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
74
+ # 2.28% MVS-Synth (synthetic, GT depth; scene 0000 held out).
75
+ - "5700 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${mvs_synth_root}', dataset_type='mvs_synth', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=5700, track_spatial_stride=1, min_overlap_ratio=0.15, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, exclude_sequence_ids=['0000']), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
76
+ # 4.52% TartanAir (static forward-moving scenes).
77
+ - "11300 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${tartanair_root}', dataset_type='tartanair', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=11300, track_spatial_stride=1, min_overlap_ratio=0.15, min_shift_ratio=0.15, max_shift_ratio=0.5, shift_beta_ab=[2.0, 5.0], src_center_in_target_prob=1.0, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/torch_profiler/rank0.1780383283598487142.pt.trace.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d4fa158ed2e4c92bf08152ca349aa5998f7b6572defa59490d8a8c6d1b6bf8e2
3
+ size 4743420315
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/debug-internal.log ADDED
@@ -0,0 +1,97 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-06-02T06:50:13.570807992Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-06-02T06:50:13.573515504Z","level":"INFO","msg":"stream: starting","core version":"0.26.0"}
3
+ {"time":"2026-06-02T06:50:14.064306628Z","level":"INFO","msg":"stream: created new stream","id":"ixxjov67"}
4
+ {"time":"2026-06-02T06:50:14.064412769Z","level":"INFO","msg":"handler: started"}
5
+ {"time":"2026-06-02T06:50:14.065276714Z","level":"INFO","msg":"stream: started"}
6
+ {"time":"2026-06-02T06:50:14.065282794Z","level":"INFO","msg":"writer: started","stream_id":"ixxjov67"}
7
+ {"time":"2026-06-02T06:50:14.06566832Z","level":"INFO","msg":"sender: started"}
8
+ {"time":"2026-06-02T06:50:15.631914228Z","level":"INFO","msg":"filestream: sending request","total_files":1,"console_offset":0,"console_lines":1}
9
+ {"time":"2026-06-02T06:50:16.106065701Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
10
+ {"time":"2026-06-02T06:50:23.169926895Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":44}
11
+ {"time":"2026-06-02T06:50:23.18417797Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":175}
12
+ {"time":"2026-06-02T06:50:30.632687583Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":0,"events_lines":2,"console_offset":1,"console_lines":126,"uploaded_len":2}
13
+ {"time":"2026-06-02T06:50:31.057643889Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
14
+ {"time":"2026-06-02T06:50:45.633956602Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":2,"events_lines":2}
15
+ {"time":"2026-06-02T06:50:45.995092564Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
16
+ {"time":"2026-06-02T06:51:00.632732136Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":4,"events_lines":2}
17
+ {"time":"2026-06-02T06:51:01.013164986Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
18
+ {"time":"2026-06-02T06:51:15.632693827Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":6,"events_lines":2}
19
+ {"time":"2026-06-02T06:51:15.973083783Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
20
+ {"time":"2026-06-02T06:51:30.632614271Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":8,"events_lines":2}
21
+ {"time":"2026-06-02T06:51:30.902457359Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
22
+ {"time":"2026-06-02T06:51:45.632638154Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":10,"events_lines":2,"console_offset":127,"console_lines":3}
23
+ {"time":"2026-06-02T06:51:46.012081416Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
24
+ {"time":"2026-06-02T06:52:00.632782948Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":0,"history_lines":1,"events_offset":12,"events_lines":2,"console_offset":126,"console_lines":1}
25
+ {"time":"2026-06-02T06:52:01.245405887Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
26
+ {"time":"2026-06-02T06:52:15.633245252Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":1,"history_lines":1,"events_offset":14,"events_lines":2,"console_offset":126,"console_lines":1}
27
+ {"time":"2026-06-02T06:52:16.031118944Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
28
+ {"time":"2026-06-02T06:52:30.632460243Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":2,"history_lines":2,"events_offset":16,"events_lines":2,"console_offset":126,"console_lines":1}
29
+ {"time":"2026-06-02T06:52:31.064836975Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
30
+ {"time":"2026-06-02T06:52:45.632586315Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":4,"history_lines":1,"events_offset":18,"events_lines":2,"console_offset":126,"console_lines":1}
31
+ {"time":"2026-06-02T06:52:45.920833256Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
32
+ {"time":"2026-06-02T06:53:00.632560879Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":5,"history_lines":1,"events_offset":20,"events_lines":2,"console_offset":126,"console_lines":1}
33
+ {"time":"2026-06-02T06:53:01.114903023Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
34
+ {"time":"2026-06-02T06:53:15.633256291Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":6,"history_lines":1,"events_offset":22,"events_lines":2,"console_offset":126,"console_lines":1}
35
+ {"time":"2026-06-02T06:53:16.048155401Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
36
+ {"time":"2026-06-02T06:53:30.633585048Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":7,"history_lines":1,"events_offset":24,"events_lines":2,"console_offset":126,"console_lines":1}
37
+ {"time":"2026-06-02T06:53:31.116840996Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
38
+ {"time":"2026-06-02T06:53:45.632811032Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":8,"history_lines":1,"events_offset":26,"events_lines":2,"console_offset":126,"console_lines":1}
39
+ {"time":"2026-06-02T06:53:45.995726219Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
40
+ {"time":"2026-06-02T06:54:00.632163343Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":28,"events_lines":2,"console_offset":130,"console_lines":1}
41
+ {"time":"2026-06-02T06:54:01.04412747Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
42
+ {"time":"2026-06-02T06:54:15.632919183Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":30,"events_lines":2}
43
+ {"time":"2026-06-02T06:54:16.033687178Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
44
+ {"time":"2026-06-02T06:54:30.632581426Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":32,"events_lines":2}
45
+ {"time":"2026-06-02T06:54:31.115164341Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
46
+ {"time":"2026-06-02T06:54:45.631988187Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":34,"events_lines":2}
47
+ {"time":"2026-06-02T06:54:46.16067707Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
48
+ {"time":"2026-06-02T06:55:00.632478441Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":36,"events_lines":2}
49
+ {"time":"2026-06-02T06:55:01.137467668Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
50
+ {"time":"2026-06-02T06:55:15.634021135Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":38,"events_lines":2}
51
+ {"time":"2026-06-02T06:55:16.528050934Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
52
+ {"time":"2026-06-02T06:55:30.633071312Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":9,"history_lines":1,"events_offset":40,"events_lines":2,"console_offset":126,"console_lines":1}
53
+ {"time":"2026-06-02T06:55:30.990818571Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
54
+ {"time":"2026-06-02T06:55:45.632948623Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":10,"history_lines":1,"events_offset":42,"events_lines":2,"console_offset":126,"console_lines":1}
55
+ {"time":"2026-06-02T06:55:45.947339192Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
56
+ {"time":"2026-06-02T06:56:00.63331661Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":11,"history_lines":1,"events_offset":44,"events_lines":2,"console_offset":126,"console_lines":1}
57
+ {"time":"2026-06-02T06:56:01.090740183Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
58
+ {"time":"2026-06-02T06:56:15.633357213Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":12,"history_lines":1,"events_offset":46,"events_lines":2,"console_offset":126,"console_lines":1}
59
+ {"time":"2026-06-02T06:56:16.100888881Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
60
+ {"time":"2026-06-02T06:56:30.633508896Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":13,"history_lines":2,"events_offset":48,"events_lines":2,"console_offset":126,"console_lines":1}
61
+ {"time":"2026-06-02T06:56:31.248935786Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
62
+ {"time":"2026-06-02T06:56:45.633656982Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":15,"history_lines":1,"events_offset":50,"events_lines":2,"console_offset":126,"console_lines":1}
63
+ {"time":"2026-06-02T06:56:46.023339526Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
64
+ {"time":"2026-06-02T06:57:00.633242778Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":16,"history_lines":1,"events_offset":52,"events_lines":2,"console_offset":126,"console_lines":1}
65
+ {"time":"2026-06-02T06:57:01.000390992Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
66
+ {"time":"2026-06-02T06:57:15.632535397Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":17,"history_lines":2,"events_offset":54,"events_lines":2,"console_offset":126,"console_lines":1}
67
+ {"time":"2026-06-02T06:57:16.015127404Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
68
+ {"time":"2026-06-02T06:57:30.633130173Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":19,"history_lines":1,"events_offset":56,"events_lines":2,"console_offset":126,"console_lines":1}
69
+ {"time":"2026-06-02T06:57:30.99356342Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
70
+ {"time":"2026-06-02T06:57:45.633329584Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":20,"history_lines":1,"events_offset":58,"events_lines":2,"console_offset":126,"console_lines":1}
71
+ {"time":"2026-06-02T06:57:46.07541001Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
72
+ {"time":"2026-06-02T06:58:00.632386559Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":21,"history_lines":1,"events_offset":60,"events_lines":2,"console_offset":126,"console_lines":1}
73
+ {"time":"2026-06-02T06:58:01.18224948Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
74
+ {"time":"2026-06-02T06:58:15.632888861Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":22,"history_lines":2,"events_offset":62,"events_lines":2,"console_offset":126,"console_lines":1}
75
+ {"time":"2026-06-02T06:58:16.030413718Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
76
+ {"time":"2026-06-02T06:58:30.633023024Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":24,"history_lines":1,"events_offset":64,"events_lines":2,"console_offset":126,"console_lines":1}
77
+ {"time":"2026-06-02T06:58:31.068636213Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
78
+ {"time":"2026-06-02T06:58:45.633331848Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":25,"history_lines":1,"events_offset":66,"events_lines":2,"console_offset":126,"console_lines":1}
79
+ {"time":"2026-06-02T06:58:46.039435816Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
80
+ {"time":"2026-06-02T06:59:00.632360803Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":26,"history_lines":1,"events_offset":68,"events_lines":2,"console_offset":126,"console_lines":1}
81
+ {"time":"2026-06-02T06:59:01.039092307Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
82
+ {"time":"2026-06-02T06:59:15.632705468Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":70,"events_lines":2}
83
+ {"time":"2026-06-02T06:59:16.085584263Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
84
+ {"time":"2026-06-02T06:59:30.632579618Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":72,"events_lines":2}
85
+ {"time":"2026-06-02T06:59:31.365655224Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
86
+ {"time":"2026-06-02T06:59:45.633048671Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":27,"history_lines":2,"events_offset":74,"events_lines":2,"console_offset":126,"console_lines":1}
87
+ {"time":"2026-06-02T06:59:46.003907075Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
88
+ {"time":"2026-06-02T07:00:00.633135873Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":29,"history_lines":1,"events_offset":76,"events_lines":2,"console_offset":126,"console_lines":1}
89
+ {"time":"2026-06-02T07:00:01.363370609Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
90
+ {"time":"2026-06-02T07:00:13.67931977Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
91
+ {"time":"2026-06-02T07:00:13.679909675Z","level":"INFO","msg":"filestream: sending request","total_files":2,"console_offset":131,"console_lines":103,"uploaded_len":3,"complete":true,"exit_code":0}
92
+ {"time":"2026-06-02T07:00:14.181094862Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
93
+ {"time":"2026-06-02T07:00:14.18215301Z","level":"INFO","msg":"handler: operation stats","stats":{}}
94
+ {"time":"2026-06-02T07:00:14.188249365Z","level":"INFO","msg":"stream: closing"}
95
+ {"time":"2026-06-02T07:00:14.188298388Z","level":"INFO","msg":"handler: closed"}
96
+ {"time":"2026-06-02T07:00:14.188708233Z","level":"INFO","msg":"sender: closed"}
97
+ {"time":"2026-06-02T07:00:14.188737768Z","level":"INFO","msg":"stream: closed"}
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/debug.log ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-06-02 06:50:12,052 INFO MainThread:152969 [wandb_setup.py:_flush():81] Current SDK version is 0.26.0
2
+ 2026-06-02 06:50:12,053 INFO MainThread:152969 [wandb_setup.py:_flush():81] Configure stats pid to 152969
3
+ 2026-06-02 06:50:12,053 INFO MainThread:152969 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-06-02 06:50:12,053 INFO MainThread:152969 [wandb_init.py:setup_run_log_directory():721] Logging user logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/logs/debug.log
5
+ 2026-06-02 06:50:12,054 INFO MainThread:152969 [wandb_init.py:setup_run_log_directory():722] Logging internal logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/logs/debug-internal.log
6
+ 2026-06-02 06:50:12,054 INFO MainThread:152969 [wandb_init.py:init():848] calling init triggers
7
+ 2026-06-02 06:50:12,060 INFO MainThread:152969 [wandb_init.py:init():853] wandb.init called with sweep_config: {}
8
+ config: {'learning_rate': 2e-05, 'weight_decay': 0.01, 'num_epochs': 200, 'save_steps': 500, 'num_workers': 7, 'prefetch_factor': 2, 'dataset_base_path': None, 'dataset_metadata_path': None, 'dataset_repeat': 1, 'dataset_num_workers': 7, 'dataloader_prefetch_factor': 2, 'data_file_keys': 'image,video', 'pointmap_stats_path': None, 'model_paths': None, 'model_id_with_origin_paths': 'Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth', 'model_variant': 'vt2v', 'extra_inputs': None, 'track_xyz_normalization': 'minmax', 'track_xyz_quantile_low': 0.02, 'track_xyz_quantile_high': 0.98, 'predict_track_residual': False, 'residual_target_xyz_source': 'clean', 'residual_zero_center_normalization': False, 'residual_linear_normalization': False, 'residual_log_scale_factor': 3.0, 'stream3r_common_scale_stat': 'mean', 'explicit_motion_loss_weight': 0.0, 'explicit_motion_loss_snr_weight': True, 'explicit_motion_loss_snr_sigmoid': True, 'explicit_motion_loss_snr_sigmoid_a': 1.0, 'explicit_motion_loss_snr_sigmoid_b': -5.0, 'explicit_motion_loss_decode_checkpoint': False, 'explicit_motion_loss_decode_chunk_size': 0, 'normalize_track_with_xyz_cond_stats': False, 'prenormalized_track_video_signed_log': False, 'use_xyz_video_condition': False, 'xyz_video_condition_scale': 1.0, 'xyz_video_condition_aug_prob': 0.0, 'xyz_video_condition_noise_std': 0.0, 'xyz_video_condition_z_noise_std': 0.0, 'xyz_video_condition_frame_scale_std': 0.0, 'xyz_video_condition_dropout_prob': 0.0, 'xyz_video_condition_mode': 'xyz', 'xyz_video_condition_cfg_dropout_prob': 0.0, 'camera_condition_cfg_dropout_prob': 0.0, 'fp8_models': None, 'offload_models': None, 'v2v_concat_mode': 'width', 'v2v_channel_patch_embed_mode': 'add', 'trainable_models': 'dit.domain_embedding,dit.control_patch_embedding', 'find_unused_parameters': False, 'task': 'sft:v2v', 'drop_text_prob': 0.1, 'drop_video_prob': 0.0, 'drop_t2v_prob': None, 'drop_i2v_prob': None, 'input_video_noise_strength': 0.0, 'input_video_noise_random_max': 500, 'output_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout', 'remove_prefix_in_ckpt': 'pipe.dit.', 'save_every_n_epochs': 2, 'max_checkpoints_to_keep': 5, 'allow_existing_checkpoints': True, 'initial_step': 2530, 'max_train_steps': 2560, 'lora_base_model': None, 'lora_target_modules': 'q,k,v,o,ffn.0,ffn.2', 'lora_rank': 32, 'lora_alpha': None, 'lora_checkpoint': None, 'preset_lora_path': None, 'preset_lora_model': None, 'use_gradient_checkpointing': True, 'use_gradient_checkpointing_offload': False, 'gradient_accumulation_steps': 1, 'use_wandb': True, 'wandb_project': 'track4d-360', 'wandb_name': 'ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp', 'wandb_entity': None, 'wandb_tags': ['ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug'], 'wandb_notes': None, 'wandb_log_gradients': False, 'wandb_log_parameters': False, 'wandb_log_freq': 100, 'wandb_watch_model': False, 'debug_latents': False, 'debug_latents_every': 500, 'debug_latents_fps': 16, 'debug_latents_quality': 5, 'debug_latents_max_frames': None, 'debug_latents_outdir': 'debug_latents', 'debug_latents_vae_tiled': True, 'debug_latents_vae_tile_size': (30, 52), 'debug_latents_vae_tile_stride': (15, 26), 'debug_missing_grads': False, 'torch_distributed_debug': None, 'use_query_frame_impulse_condition': True, 'height': 288, 'width': 512, 'max_pixels': 1048576, 'num_frames': 49, 'tokenizer_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl', 'dit_finetune_modules': 'self_attn', 'track_injection_mode': 'per_block', 'track_injection_block_mode': 'concat_project', 'train_sigma_shift': 5.0, 'dummy': False, 'steps_per_epoch': 25000, 'initialize_model_on_cpu': False, 'resume_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors', 'gs_head_init_checkpoint': None, 'joint_track_prediction': False, 'joint_track_depth_prediction': False, 'joint_depth_prediction': False, 'target_depth_log_cap': 14.0, 'target_depth_norm_type': 'log_depth', 'target_depth_max_disparity': 5.0, 'vxf_init_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors', 'channel_concat_stage': None, 'use_residual_tracks': True, 'freeze_lora_after_load': False, 'skip_v2v_required_trainables': False, 'gs_only_training': False, 'freeze_track_block_injector': False, 'freeze_track4d_condition_modules': False, 'gs_head_style': 'flashworld', 'use_gs_output': False, 'gs_zero_features': False, 'gs_feat_dim': 1024, 'gs_loss_weight': 1.0, 'gs_mse_loss_weight': 1.0, 'gs_lpips_loss_weight': 0.5, 'gs_depth_loss_weight': 0.05, 'gs_abs_depth_loss_weight': 0.0, 'gs_opacity_loss_weight': 0.1, 'gs_opacity_prune_keep_frac': 1.0, 'gs_rgb_loss_weight': None, 'gs_sh_degree': 2, 'gs_num_points_per_pixel': 2, 'no_gs_latent_detach': False, 'use_gs_feedback_training': False, 'gs_scale_cap': None, 'gs_render_patch_size': None, 'gs_legacy_union_render': False, 'gs_lyra_depth_param': 'exp', 'gs_lyra_log_depth_shift': 0.0, 'gs_lyra_dnear': 0.1, 'gs_lyra_dfar': 10.0, 'gs_replay_cache_size': 0, 'gs_dynamic_mode': 'static', 'gs_track_supervision_mode': 'pointmap', 'gs_pos_loss_weight': 1.0, 'gs_dyn_depth_loss_weight': 0.5, 'gs_opa_tv_weight': 0.01, 'gs_xyz_tv_weight': 0.01, 'gs_predict_rotation': False, 'gs_disable_delta_opacity': False, 'gs_direct_pos_loss_weight': 0.0, 'gs_track_canonical_loss_weight': 0.0, 'gs_track_delta_loss_weight': 0.0, 'gs_track_final_pos_loss_weight': 0.0, 'gs_dynamic_max_t_video': 64, 'gs_assoc_track_2d_loss_weight': 0.0, 'gs_track_accel_loss_weight': 0.0, 'gs_track_loss_quantile': 0.98, 'gs_track_canonical_detach': False, 'gs_track_joint_render': False, 'cond_drop_track_prob': 0.1, 'cond_drop_dense_prob': 0.0, 'cond_drop_geo_prob': 0.1, 'use_track_null_token_fallback': False, 'use_dense_null_token_fallback': False, 'use_plucker_camera_control': True, 'enable_v2v_plucker_camera_control': True, 'train_plucker_camera_control': False, 'use_plucker_raymap_tokens': True, 'plucker_use_ray_origin': False, 'use_perlayer_camera_control': True, 'camera_injection_mode': 'projector', 'camera_plucker_encoder_mode': 'patchify', 'warped_condition_mode': 'off', 'warped_appearance_fusion': 'concat_proj', 'warped_geom_only_dense': True, 'multi_frame_warped_appearance': False, 'multi_frame_motion_threshold': 0.01, 'fixed_step_frac': None, 'fixed_num_inference_steps': 50, 'fixed_sigma': None, 'skip_diffusion_loss': False, 'use_dense_branch': True, 'dense_in_channels': 5, 'dense_proj_dim': 32, 'dense_num_residual_blocks': 2, 'dense_alpha_track': 1.0, 'use_depth_null_token': False, 'dense_conditioning_noise_max_level': 0.15, 'source_mask_boundary_depth_dropout_prob': 0.5, 'source_mask_boundary_kernel_size': 5, 'dense_warp_mode': 'hybrid_softsplat', 'dense_warp_motion_threshold': 0.005, 'dense_hybrid_disable_dynamic_masks': False, 'use_vista4d_latent_streams': False, 'use_gen3c_warp_latent_channels': False, 'disable_track_modules': False, 'dense_only_condition_modules': False, 'torch_profile_active_steps': 5, 'torch_profile_wait_steps': 3, 'torch_profile_warmup_steps': 1, 'torch_profile_repeat': 1, 'torch_profile_dir': None, 'torch_profile_record_shapes': False, 'torch_profile_memory': False, 'torch_profile_with_stack': False, 'track_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth_noconv3dres.yaml', 'track4d360_data_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml', 'syn4d_root': '/scratch/shared/beegfs/zeren/Syn4D', 'scene': None, 'sequence_idx': None, 'track_spatial_stride': 1, '_wandb': {}}
9
+ 2026-06-02 06:50:12,060 INFO MainThread:152969 [wandb_init.py:init():896] starting backend
10
+ 2026-06-02 06:50:12,777 INFO MainThread:152969 [wandb_init.py:init():911] sending inform_init request
11
+ 2026-06-02 06:50:13,568 INFO MainThread:152969 [wandb_init.py:init():919] backend started and connected
12
+ 2026-06-02 06:50:13,571 INFO MainThread:152969 [wandb_init.py:init():989] updated telemetry
13
+ 2026-06-02 06:50:13,661 INFO MainThread:152969 [wandb_init.py:init():1013] communicating run to backend with 90.0 second timeout
14
+ 2026-06-02 06:50:14,653 INFO MainThread:152969 [wandb_init.py:init():1058] starting run threads in backend
15
+ 2026-06-02 06:50:15,600 INFO MainThread:152969 [wandb_run.py:_console_start():2542] atexit reg
16
+ 2026-06-02 06:50:15,600 INFO MainThread:152969 [wandb_run.py:_redirect():2391] redirect: wrap_raw
17
+ 2026-06-02 06:50:15,600 INFO MainThread:152969 [wandb_run.py:_redirect():2460] Wrapping output streams.
18
+ 2026-06-02 06:50:15,600 INFO MainThread:152969 [wandb_run.py:_redirect():2483] Redirects installed.
19
+ 2026-06-02 06:50:15,631 INFO MainThread:152969 [wandb_init.py:init():1098] run started, returning control to user process
20
+ 2026-06-02 07:00:12,998 INFO MainThread:152969 [wandb_run.py:_finish():2309] finishing run nirvanalan/track4d-360/ixxjov67
21
+ 2026-06-02 07:00:13,001 INFO MainThread:152969 [wandb_run.py:_atexit_cleanup():2508] got exitcode: 0
22
+ 2026-06-02 07:00:13,002 INFO MainThread:152969 [wandb_run.py:_restore():2490] restore
23
+ 2026-06-02 07:00:13,002 INFO MainThread:152969 [wandb_run.py:_restore():2496] restore done
24
+ 2026-06-02 07:00:14,187 INFO MainThread:152969 [wandb_run.py:_footer_sync_info():3886] logging synced files
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/files/config.yaml ADDED
@@ -0,0 +1,700 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _wandb:
2
+ value:
3
+ cli_version: 0.26.0
4
+ e:
5
+ 1m9b4etgqyfvjqvydz51tku5m7qgycq3:
6
+ args:
7
+ - --model_id_with_origin_paths
8
+ - Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth
9
+ - --tokenizer_path
10
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl
11
+ - --task
12
+ - sft:v2v
13
+ - --model_variant
14
+ - vt2v
15
+ - --v2v_concat_mode
16
+ - width
17
+ - --height
18
+ - "288"
19
+ - --width
20
+ - "512"
21
+ - --num_frames
22
+ - "49"
23
+ - --track4d360_data_config
24
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml
25
+ - --steps_per_epoch
26
+ - "25000"
27
+ - --num_epochs
28
+ - "200"
29
+ - --learning_rate
30
+ - "2e-5"
31
+ - --gradient_accumulation_steps
32
+ - "1"
33
+ - --save_steps
34
+ - "500"
35
+ - --remove_prefix_in_ckpt
36
+ - pipe.dit.
37
+ - --trainable_models
38
+ - dit.domain_embedding,dit.control_patch_embedding
39
+ - --dit_finetune_modules
40
+ - self_attn
41
+ - --track_injection_mode
42
+ - per_block
43
+ - --track_injection_block_mode
44
+ - concat_project
45
+ - --track_config
46
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth_noconv3dres.yaml
47
+ - --train_sigma_shift
48
+ - "5.0"
49
+ - --use_query_frame_impulse_condition
50
+ - --use_plucker_camera_control
51
+ - --enable_v2v_plucker_camera_control
52
+ - --use_plucker_raymap_tokens
53
+ - --use_perlayer_camera_control
54
+ - --camera_injection_mode
55
+ - projector
56
+ - --camera_plucker_encoder_mode
57
+ - patchify
58
+ - --use_dense_branch
59
+ - --dense_in_channels
60
+ - "5"
61
+ - --dense_proj_dim
62
+ - "32"
63
+ - --dense_num_residual_blocks
64
+ - "2"
65
+ - --dense_alpha_track
66
+ - "1.0"
67
+ - --dense_conditioning_noise_max_level
68
+ - "0.15"
69
+ - --source_mask_boundary_depth_dropout_prob
70
+ - "0.5"
71
+ - --source_mask_boundary_kernel_size
72
+ - "5"
73
+ - --dense_warp_mode
74
+ - hybrid_softsplat
75
+ - --dense_warp_motion_threshold
76
+ - "0.005"
77
+ - --cond_drop_geo_prob
78
+ - "0.1"
79
+ - --cond_drop_track_prob
80
+ - "0.1"
81
+ - --cond_drop_dense_prob
82
+ - "0.0"
83
+ - --initial_step
84
+ - "2530"
85
+ - --output_path
86
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout
87
+ - --dataset_num_workers
88
+ - "7"
89
+ - --dataloader_prefetch_factor
90
+ - "2"
91
+ - --drop_text_prob
92
+ - "0.1"
93
+ - --debug_latents_every
94
+ - "500"
95
+ - --allow_existing_checkpoints
96
+ - --wandb_project
97
+ - track4d-360
98
+ - --wandb_name
99
+ - ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp
100
+ - --wandb_tags
101
+ - ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug
102
+ - --use_wandb
103
+ - --vxf_init_checkpoint
104
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors
105
+ - --use_gradient_checkpointing
106
+ - --resume_checkpoint
107
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors
108
+ - --max_train_steps
109
+ - "2560"
110
+ - --torch_profile_active_steps
111
+ - "5"
112
+ - --torch_profile_wait_steps
113
+ - "3"
114
+ - --torch_profile_warmup_steps
115
+ - "1"
116
+ - --torch_profile_repeat
117
+ - "1"
118
+ - --no_torch_profile_record_shapes
119
+ - --no_torch_profile_memory
120
+ codePath: scripts/wan2.1_fun/train_track4d360.py
121
+ codePathLocal: scripts/wan2.1_fun/train_track4d360.py
122
+ cpu_count: 288
123
+ cpu_count_logical: 288
124
+ cudaVersion: "12.7"
125
+ disk:
126
+ /:
127
+ total: "255795724288"
128
+ used: "546373632"
129
+ email: yushi.lan@outlook.com
130
+ executable: /projects/public/u6ew/yslan_public/yslan_project/miniforge3/envs/th29py312/bin/python3.12
131
+ git:
132
+ commit: b6681ed2127310d14bc5b93ae5f1d20df8b9e907
133
+ remote: git@github.com:NIRVANALAN/track4d-360.git
134
+ gpu: NVIDIA GH200 120GB
135
+ gpu_count: 4
136
+ gpu_nvidia:
137
+ - architecture: Hopper
138
+ cudaCores: 16896
139
+ memoryTotal: "102625181696"
140
+ name: NVIDIA GH200 120GB
141
+ uuid: GPU-fda2ad82-f650-d1df-fbce-bab2bc8dd421
142
+ - architecture: Hopper
143
+ cudaCores: 16896
144
+ memoryTotal: "102625181696"
145
+ name: NVIDIA GH200 120GB
146
+ uuid: GPU-f29ba6f4-4da2-cc4b-37dc-d260fa79a222
147
+ - architecture: Hopper
148
+ cudaCores: 16896
149
+ memoryTotal: "102625181696"
150
+ name: NVIDIA GH200 120GB
151
+ uuid: GPU-38da6d83-51b2-5586-2248-c681407f2a51
152
+ - architecture: Hopper
153
+ cudaCores: 16896
154
+ memoryTotal: "102625181696"
155
+ name: NVIDIA GH200 120GB
156
+ uuid: GPU-aa24fa01-5dca-c359-28f7-7adf38ef534a
157
+ host: nid010301
158
+ memory:
159
+ total: "919613407232"
160
+ os: Linux-6.4.0-150600.23.73_15.0.14-cray_shasta_c_64k-aarch64-with-glibc2.38
161
+ program: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/scripts/wan2.1_fun/train_track4d360.py
162
+ python: CPython 3.12.13
163
+ root: /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout
164
+ slurm:
165
+ cluster_name: gracehopper
166
+ conf: /var/spool/slurmd/conf-cache/slurm.conf
167
+ cpu_bind: quiet,mask_cpu:0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
168
+ cpu_bind_list: 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
169
+ cpu_bind_type: 'mask_cpu:'
170
+ cpu_bind_verbose: quiet
171
+ cpus_on_node: "288"
172
+ distribution: cyclic
173
+ gpus: "32"
174
+ gpus_on_node: "4"
175
+ gpus_per_node: "4"
176
+ gtids: "0"
177
+ job_account: brics.u6ew
178
+ job_cpus_per_node: 288(x8)
179
+ job_end_time: "1780389885"
180
+ job_gid: "1483804847"
181
+ job_gpus: 0,1,2,3
182
+ job_id: "4982942"
183
+ job_name: 14b_pair1x49_profile_ablate_conv3dres0_fsdp2
184
+ job_nodelist: nid[010301-010304,010306,010308-010309,010321]
185
+ job_num_nodes: "8"
186
+ job_partition: workq
187
+ job_qos: normal
188
+ job_start_time: "1780382685"
189
+ job_uid: "1483804847"
190
+ job_user: yushi.u6ew
191
+ jobid: "4982942"
192
+ launch_node_ipaddr: 10.242.5.140
193
+ localid: "0"
194
+ mpi_type: pmi2
195
+ nnodes: "8"
196
+ nodeid: "0"
197
+ nodelist: nid[010301-010304,010306,010308-010309,010321]
198
+ nprocs: "8"
199
+ ntasks: "8"
200
+ ntasks_per_node: "1"
201
+ oom_kill_step: "0"
202
+ prio_process: "0"
203
+ procid: "0"
204
+ srun_comm_host: 10.242.5.140
205
+ srun_comm_port: "45579"
206
+ step_gpus: 0,1,2,3
207
+ step_id: "0"
208
+ step_launcher_port: "45579"
209
+ step_nodelist: nid[010301-010304,010306,010308-010309,010321]
210
+ step_num_nodes: "8"
211
+ step_num_tasks: "8"
212
+ step_resv_ports: 26618-26619
213
+ step_tasks_per_node: 1(x8)
214
+ stepid: "0"
215
+ submit_dir: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360
216
+ submit_host: login40
217
+ task_pid: "152888"
218
+ tasks_per_node: 1(x8)
219
+ topology_addr: global.group4.nid010301
220
+ topology_addr_pattern: switch.switch.node
221
+ umask: "0022"
222
+ startedAt: "2026-06-02T06:50:12.043060Z"
223
+ writerId: 1m9b4etgqyfvjqvydz51tku5m7qgycq3
224
+ m: []
225
+ python_version: 3.12.13
226
+ t:
227
+ "1":
228
+ - 1
229
+ - 11
230
+ - 41
231
+ - 49
232
+ - 71
233
+ - 83
234
+ - 98
235
+ - 106
236
+ "2":
237
+ - 1
238
+ - 11
239
+ - 41
240
+ - 49
241
+ - 51
242
+ - 71
243
+ - 83
244
+ - 98
245
+ - 106
246
+ "3":
247
+ - 2
248
+ - 13
249
+ - 15
250
+ - 16
251
+ - 41
252
+ - 61
253
+ "4": 3.12.13
254
+ "5": 0.26.0
255
+ "6": 4.46.2
256
+ "12": 0.26.0
257
+ "13": linux-aarch64
258
+ allow_existing_checkpoints:
259
+ value: true
260
+ camera_condition_cfg_dropout_prob:
261
+ value: 0
262
+ camera_injection_mode:
263
+ value: projector
264
+ camera_plucker_encoder_mode:
265
+ value: patchify
266
+ channel_concat_stage:
267
+ value: null
268
+ cond_drop_dense_prob:
269
+ value: 0
270
+ cond_drop_geo_prob:
271
+ value: 0.1
272
+ cond_drop_track_prob:
273
+ value: 0.1
274
+ data_file_keys:
275
+ value: image,video
276
+ dataloader_prefetch_factor:
277
+ value: 2
278
+ dataset_base_path:
279
+ value: null
280
+ dataset_metadata_path:
281
+ value: null
282
+ dataset_num_workers:
283
+ value: 7
284
+ dataset_repeat:
285
+ value: 1
286
+ debug_latents:
287
+ value: false
288
+ debug_latents_every:
289
+ value: 500
290
+ debug_latents_fps:
291
+ value: 16
292
+ debug_latents_max_frames:
293
+ value: null
294
+ debug_latents_outdir:
295
+ value: debug_latents
296
+ debug_latents_quality:
297
+ value: 5
298
+ debug_latents_vae_tile_size:
299
+ value:
300
+ - 30
301
+ - 52
302
+ debug_latents_vae_tile_stride:
303
+ value:
304
+ - 15
305
+ - 26
306
+ debug_latents_vae_tiled:
307
+ value: true
308
+ debug_missing_grads:
309
+ value: false
310
+ dense_alpha_track:
311
+ value: 1
312
+ dense_conditioning_noise_max_level:
313
+ value: 0.15
314
+ dense_hybrid_disable_dynamic_masks:
315
+ value: false
316
+ dense_in_channels:
317
+ value: 5
318
+ dense_num_residual_blocks:
319
+ value: 2
320
+ dense_only_condition_modules:
321
+ value: false
322
+ dense_proj_dim:
323
+ value: 32
324
+ dense_warp_mode:
325
+ value: hybrid_softsplat
326
+ dense_warp_motion_threshold:
327
+ value: 0.005
328
+ disable_track_modules:
329
+ value: false
330
+ dit_finetune_modules:
331
+ value: self_attn
332
+ drop_i2v_prob:
333
+ value: null
334
+ drop_t2v_prob:
335
+ value: null
336
+ drop_text_prob:
337
+ value: 0.1
338
+ drop_video_prob:
339
+ value: 0
340
+ dummy:
341
+ value: false
342
+ enable_v2v_plucker_camera_control:
343
+ value: true
344
+ explicit_motion_loss_decode_checkpoint:
345
+ value: false
346
+ explicit_motion_loss_decode_chunk_size:
347
+ value: 0
348
+ explicit_motion_loss_snr_sigmoid:
349
+ value: true
350
+ explicit_motion_loss_snr_sigmoid_a:
351
+ value: 1
352
+ explicit_motion_loss_snr_sigmoid_b:
353
+ value: -5
354
+ explicit_motion_loss_snr_weight:
355
+ value: true
356
+ explicit_motion_loss_weight:
357
+ value: 0
358
+ extra_inputs:
359
+ value: null
360
+ find_unused_parameters:
361
+ value: false
362
+ fixed_num_inference_steps:
363
+ value: 50
364
+ fixed_sigma:
365
+ value: null
366
+ fixed_step_frac:
367
+ value: null
368
+ fp8_models:
369
+ value: null
370
+ freeze_lora_after_load:
371
+ value: false
372
+ freeze_track_block_injector:
373
+ value: false
374
+ freeze_track4d_condition_modules:
375
+ value: false
376
+ gradient_accumulation_steps:
377
+ value: 1
378
+ gs_abs_depth_loss_weight:
379
+ value: 0
380
+ gs_assoc_track_2d_loss_weight:
381
+ value: 0
382
+ gs_depth_loss_weight:
383
+ value: 0.05
384
+ gs_direct_pos_loss_weight:
385
+ value: 0
386
+ gs_disable_delta_opacity:
387
+ value: false
388
+ gs_dyn_depth_loss_weight:
389
+ value: 0.5
390
+ gs_dynamic_max_t_video:
391
+ value: 64
392
+ gs_dynamic_mode:
393
+ value: static
394
+ gs_feat_dim:
395
+ value: 1024
396
+ gs_head_init_checkpoint:
397
+ value: null
398
+ gs_head_style:
399
+ value: flashworld
400
+ gs_legacy_union_render:
401
+ value: false
402
+ gs_loss_weight:
403
+ value: 1
404
+ gs_lpips_loss_weight:
405
+ value: 0.5
406
+ gs_lyra_depth_param:
407
+ value: exp
408
+ gs_lyra_dfar:
409
+ value: 10
410
+ gs_lyra_dnear:
411
+ value: 0.1
412
+ gs_lyra_log_depth_shift:
413
+ value: 0
414
+ gs_mse_loss_weight:
415
+ value: 1
416
+ gs_num_points_per_pixel:
417
+ value: 2
418
+ gs_only_training:
419
+ value: false
420
+ gs_opa_tv_weight:
421
+ value: 0.01
422
+ gs_opacity_loss_weight:
423
+ value: 0.1
424
+ gs_opacity_prune_keep_frac:
425
+ value: 1
426
+ gs_pos_loss_weight:
427
+ value: 1
428
+ gs_predict_rotation:
429
+ value: false
430
+ gs_render_patch_size:
431
+ value: null
432
+ gs_replay_cache_size:
433
+ value: 0
434
+ gs_rgb_loss_weight:
435
+ value: null
436
+ gs_scale_cap:
437
+ value: null
438
+ gs_sh_degree:
439
+ value: 2
440
+ gs_track_accel_loss_weight:
441
+ value: 0
442
+ gs_track_canonical_detach:
443
+ value: false
444
+ gs_track_canonical_loss_weight:
445
+ value: 0
446
+ gs_track_delta_loss_weight:
447
+ value: 0
448
+ gs_track_final_pos_loss_weight:
449
+ value: 0
450
+ gs_track_joint_render:
451
+ value: false
452
+ gs_track_loss_quantile:
453
+ value: 0.98
454
+ gs_track_supervision_mode:
455
+ value: pointmap
456
+ gs_xyz_tv_weight:
457
+ value: 0.01
458
+ gs_zero_features:
459
+ value: false
460
+ height:
461
+ value: 288
462
+ initial_step:
463
+ value: 2530
464
+ initialize_model_on_cpu:
465
+ value: false
466
+ input_video_noise_random_max:
467
+ value: 500
468
+ input_video_noise_strength:
469
+ value: 0
470
+ joint_depth_prediction:
471
+ value: false
472
+ joint_track_depth_prediction:
473
+ value: false
474
+ joint_track_prediction:
475
+ value: false
476
+ learning_rate:
477
+ value: 2e-05
478
+ lora_alpha:
479
+ value: null
480
+ lora_base_model:
481
+ value: null
482
+ lora_checkpoint:
483
+ value: null
484
+ lora_rank:
485
+ value: 32
486
+ lora_target_modules:
487
+ value: q,k,v,o,ffn.0,ffn.2
488
+ max_checkpoints_to_keep:
489
+ value: 5
490
+ max_pixels:
491
+ value: 1048576
492
+ max_train_steps:
493
+ value: 2560
494
+ model_id_with_origin_paths:
495
+ value: Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth
496
+ model_paths:
497
+ value: null
498
+ model_variant:
499
+ value: vt2v
500
+ multi_frame_motion_threshold:
501
+ value: 0.01
502
+ multi_frame_warped_appearance:
503
+ value: false
504
+ no_gs_latent_detach:
505
+ value: false
506
+ normalize_track_with_xyz_cond_stats:
507
+ value: false
508
+ num_epochs:
509
+ value: 200
510
+ num_frames:
511
+ value: 49
512
+ num_workers:
513
+ value: 7
514
+ offload_models:
515
+ value: null
516
+ output_path:
517
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout
518
+ plucker_use_ray_origin:
519
+ value: false
520
+ pointmap_stats_path:
521
+ value: null
522
+ predict_track_residual:
523
+ value: false
524
+ prefetch_factor:
525
+ value: 2
526
+ prenormalized_track_video_signed_log:
527
+ value: false
528
+ preset_lora_model:
529
+ value: null
530
+ preset_lora_path:
531
+ value: null
532
+ remove_prefix_in_ckpt:
533
+ value: pipe.dit.
534
+ residual_linear_normalization:
535
+ value: false
536
+ residual_log_scale_factor:
537
+ value: 3
538
+ residual_target_xyz_source:
539
+ value: clean
540
+ residual_zero_center_normalization:
541
+ value: false
542
+ resume_checkpoint:
543
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors
544
+ save_every_n_epochs:
545
+ value: 2
546
+ save_steps:
547
+ value: 500
548
+ scene:
549
+ value: null
550
+ sequence_idx:
551
+ value: null
552
+ skip_diffusion_loss:
553
+ value: false
554
+ skip_v2v_required_trainables:
555
+ value: false
556
+ source_mask_boundary_depth_dropout_prob:
557
+ value: 0.5
558
+ source_mask_boundary_kernel_size:
559
+ value: 5
560
+ steps_per_epoch:
561
+ value: 25000
562
+ stream3r_common_scale_stat:
563
+ value: mean
564
+ syn4d_root:
565
+ value: /scratch/shared/beegfs/zeren/Syn4D
566
+ target_depth_log_cap:
567
+ value: 14
568
+ target_depth_max_disparity:
569
+ value: 5
570
+ target_depth_norm_type:
571
+ value: log_depth
572
+ task:
573
+ value: sft:v2v
574
+ tokenizer_path:
575
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl
576
+ torch_distributed_debug:
577
+ value: null
578
+ torch_profile_active_steps:
579
+ value: 5
580
+ torch_profile_dir:
581
+ value: null
582
+ torch_profile_memory:
583
+ value: false
584
+ torch_profile_record_shapes:
585
+ value: false
586
+ torch_profile_repeat:
587
+ value: 1
588
+ torch_profile_wait_steps:
589
+ value: 3
590
+ torch_profile_warmup_steps:
591
+ value: 1
592
+ torch_profile_with_stack:
593
+ value: false
594
+ track_config:
595
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth_noconv3dres.yaml
596
+ track_injection_block_mode:
597
+ value: concat_project
598
+ track_injection_mode:
599
+ value: per_block
600
+ track_spatial_stride:
601
+ value: 1
602
+ track_xyz_normalization:
603
+ value: minmax
604
+ track_xyz_quantile_high:
605
+ value: 0.98
606
+ track_xyz_quantile_low:
607
+ value: 0.02
608
+ track4d360_data_config:
609
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml
610
+ train_plucker_camera_control:
611
+ value: false
612
+ train_sigma_shift:
613
+ value: 5
614
+ trainable_models:
615
+ value: dit.domain_embedding,dit.control_patch_embedding
616
+ use_dense_branch:
617
+ value: true
618
+ use_dense_null_token_fallback:
619
+ value: false
620
+ use_depth_null_token:
621
+ value: false
622
+ use_gen3c_warp_latent_channels:
623
+ value: false
624
+ use_gradient_checkpointing:
625
+ value: true
626
+ use_gradient_checkpointing_offload:
627
+ value: false
628
+ use_gs_feedback_training:
629
+ value: false
630
+ use_gs_output:
631
+ value: false
632
+ use_perlayer_camera_control:
633
+ value: true
634
+ use_plucker_camera_control:
635
+ value: true
636
+ use_plucker_raymap_tokens:
637
+ value: true
638
+ use_query_frame_impulse_condition:
639
+ value: true
640
+ use_residual_tracks:
641
+ value: true
642
+ use_track_null_token_fallback:
643
+ value: false
644
+ use_vista4d_latent_streams:
645
+ value: false
646
+ use_wandb:
647
+ value: true
648
+ use_xyz_video_condition:
649
+ value: false
650
+ v2v_channel_patch_embed_mode:
651
+ value: add
652
+ v2v_concat_mode:
653
+ value: width
654
+ vxf_init_checkpoint:
655
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors
656
+ wandb_entity:
657
+ value: null
658
+ wandb_log_freq:
659
+ value: 100
660
+ wandb_log_gradients:
661
+ value: false
662
+ wandb_log_parameters:
663
+ value: false
664
+ wandb_name:
665
+ value: ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp
666
+ wandb_notes:
667
+ value: null
668
+ wandb_project:
669
+ value: track4d-360
670
+ wandb_tags:
671
+ value:
672
+ - ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug
673
+ wandb_watch_model:
674
+ value: false
675
+ warped_appearance_fusion:
676
+ value: concat_proj
677
+ warped_condition_mode:
678
+ value: "off"
679
+ warped_geom_only_dense:
680
+ value: true
681
+ weight_decay:
682
+ value: 0.01
683
+ width:
684
+ value: 512
685
+ xyz_video_condition_aug_prob:
686
+ value: 0
687
+ xyz_video_condition_cfg_dropout_prob:
688
+ value: 0
689
+ xyz_video_condition_dropout_prob:
690
+ value: 0
691
+ xyz_video_condition_frame_scale_std:
692
+ value: 0
693
+ xyz_video_condition_mode:
694
+ value: xyz
695
+ xyz_video_condition_noise_std:
696
+ value: 0
697
+ xyz_video_condition_scale:
698
+ value: 1
699
+ xyz_video_condition_z_noise_std:
700
+ value: 0
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/files/output.log ADDED
@@ -0,0 +1,234 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ✓ WandB initialized: https://wandb.ai/nirvanalan/track4d-360/runs/ixxjov67
2
+ /projects/public/u6ew/yslan_public/yslan_project/miniforge3/envs/th29py312/lib/python3.12/site-packages/accelerate/utils/fsdp_utils.py:742: UserWarning: FSDP upcast of low precision parameters to fp32 (since mixed_precision != 'no') may affect the precision of model checkpoints. This effects 593 parameters: ['pipe.dit.domain_embedding', 'pipe.dit.blocks.0.self_attn.q.weight', 'pipe.dit.blocks.0.self_attn.q.bias', 'pipe.dit.blocks.0.self_attn.k.weight', 'pipe.dit.blocks.0.self_attn.k.bias', 'pipe.dit.blocks.0.self_attn.v.weight', 'pipe.dit.blocks.0.self_attn.v.bias', 'pipe.dit.blocks.0.self_attn.o.weight', 'pipe.dit.blocks.0.self_attn.o.bias', 'pipe.dit.blocks.0.self_attn.norm_q.weight', 'pipe.dit.blocks.0.self_attn.norm_k.weight', 'pipe.dit.blocks.0.camera_projector.weight', 'pipe.dit.blocks.0.camera_projector.bias', 'pipe.dit.blocks.1.self_attn.q.weight', 'pipe.dit.blocks.1.self_attn.q.bias', 'pipe.dit.blocks.1.self_attn.k.weight', 'pipe.dit.blocks.1.self_attn.k.bias', 'pipe.dit.blocks.1.self_attn.v.weight', 'pipe.dit.blocks.1.self_attn.v.bias', 'pipe.dit.blocks.1.self_attn.o.weight', 'pipe.dit.blocks.1.self_attn.o.bias', 'pipe.dit.blocks.1.self_attn.norm_q.weight', 'pipe.dit.blocks.1.self_attn.norm_k.weight', 'pipe.dit.blocks.1.camera_projector.weight', 'pipe.dit.blocks.1.camera_projector.bias', 'pipe.dit.blocks.2.self_attn.q.weight', 'pipe.dit.blocks.2.self_attn.q.bias', 'pipe.dit.blocks.2.self_attn.k.weight', 'pipe.dit.blocks.2.self_attn.k.bias', 'pipe.dit.blocks.2.self_attn.v.weight', 'pipe.dit.blocks.2.self_attn.v.bias', 'pipe.dit.blocks.2.self_attn.o.weight', 'pipe.dit.blocks.2.self_attn.o.bias', 'pipe.dit.blocks.2.self_attn.norm_q.weight', 'pipe.dit.blocks.2.self_attn.norm_k.weight', 'pipe.dit.blocks.2.camera_projector.weight', 'pipe.dit.blocks.2.camera_projector.bias', 'pipe.dit.blocks.3.self_attn.q.weight', 'pipe.dit.blocks.3.self_attn.q.bias', 'pipe.dit.blocks.3.self_attn.k.weight', 'pipe.dit.blocks.3.self_attn.k.bias', 'pipe.dit.blocks.3.self_attn.v.weight', 'pipe.dit.blocks.3.self_attn.v.bias', 'pipe.dit.blocks.3.self_attn.o.weight', 'pipe.dit.blocks.3.self_attn.o.bias', 'pipe.dit.blocks.3.self_attn.norm_q.weight', 'pipe.dit.blocks.3.self_attn.norm_k.weight', 'pipe.dit.blocks.3.camera_projector.weight', 'pipe.dit.blocks.3.camera_projector.bias', 'pipe.dit.blocks.4.self_attn.q.weight', 'pipe.dit.blocks.4.self_attn.q.bias', 'pipe.dit.blocks.4.self_attn.k.weight', 'pipe.dit.blocks.4.self_attn.k.bias', 'pipe.dit.blocks.4.self_attn.v.weight', 'pipe.dit.blocks.4.self_attn.v.bias', 'pipe.dit.blocks.4.self_attn.o.weight', 'pipe.dit.blocks.4.self_attn.o.bias', 'pipe.dit.blocks.4.self_attn.norm_q.weight', 'pipe.dit.blocks.4.self_attn.norm_k.weight', 'pipe.dit.blocks.4.camera_projector.weight', 'pipe.dit.blocks.4.camera_projector.bias', 'pipe.dit.blocks.5.self_attn.q.weight', 'pipe.dit.blocks.5.self_attn.q.bias', 'pipe.dit.blocks.5.self_attn.k.weight', 'pipe.dit.blocks.5.self_attn.k.bias', 'pipe.dit.blocks.5.self_attn.v.weight', 'pipe.dit.blocks.5.self_attn.v.bias', 'pipe.dit.blocks.5.self_attn.o.weight', 'pipe.dit.blocks.5.self_attn.o.bias', 'pipe.dit.blocks.5.self_attn.norm_q.weight', 'pipe.dit.blocks.5.self_attn.norm_k.weight', 'pipe.dit.blocks.5.camera_projector.weight', 'pipe.dit.blocks.5.camera_projector.bias', 'pipe.dit.blocks.6.self_attn.q.weight', 'pipe.dit.blocks.6.self_attn.q.bias', 'pipe.dit.blocks.6.self_attn.k.weight', 'pipe.dit.blocks.6.self_attn.k.bias', 'pipe.dit.blocks.6.self_attn.v.weight', 'pipe.dit.blocks.6.self_attn.v.bias', 'pipe.dit.blocks.6.self_attn.o.weight', 'pipe.dit.blocks.6.self_attn.o.bias', 'pipe.dit.blocks.6.self_attn.norm_q.weight', 'pipe.dit.blocks.6.self_attn.norm_k.weight', 'pipe.dit.blocks.6.camera_projector.weight', 'pipe.dit.blocks.6.camera_projector.bias', 'pipe.dit.blocks.7.self_attn.q.weight', 'pipe.dit.blocks.7.self_attn.q.bias', 'pipe.dit.blocks.7.self_attn.k.weight', 'pipe.dit.blocks.7.self_attn.k.bias', 'pipe.dit.blocks.7.self_attn.v.weight', 'pipe.dit.blocks.7.self_attn.v.bias', 'pipe.dit.blocks.7.self_attn.o.weight', 'pipe.dit.blocks.7.self_attn.o.bias', 'pipe.dit.block
3
+ warnings.warn(
4
+
5
+ ================================================================================
6
+ TRAINABLE PARAMETERS VERIFICATION
7
+ ================================================================================
8
+
9
+ Trainable modules (94 module groups):
10
+
11
+ 📦 pipe.dit.blocks.* — 40 blocks [0..39]:
12
+ Params per block: 131,107,840 × 40 = 5,244,313,600
13
+ Submodules (example from block 0):
14
+ • self_attn: 104,888,320
15
+ • camera_projector: 26,219,520
16
+
17
+ 📦 pipe.dense_target_control_encoder.conv3d_patchify:
18
+ Total params: 10,272
19
+ ├─ weight: 10,240
20
+ ├─ bias: 32
21
+
22
+ 📦 pipe.dense_target_control_encoder.null_token:
23
+ Total params: 156
24
+ pipe.dense_target_control_encoder.null_token: 156
25
+
26
+ 📦 pipe.dense_target_control_encoder.proj_up:
27
+ Total params: 5,148
28
+ ├─ weight: 4,992
29
+ ├─ bias: 156
30
+
31
+ 📦 pipe.dense_target_control_encoder.residual_blocks:
32
+ Total params: 110,720
33
+ ├─ 0.conv1.weight: 27,648
34
+ ├─ 0.conv2.weight: 27,648
35
+ ├─ 1.conv1.weight: 27,648
36
+ ├─ 1.conv2.weight: 27,648
37
+ ├─ 0.conv1.bias: 32
38
+ ├─ 0.conv2.bias: 32
39
+ ├─ 1.conv1.bias: 32
40
+ ├─ 1.conv2.bias: 32
41
+
42
+ 📦 pipe.dit._camera_patchify_conv:
43
+ Total params: 36,870
44
+ ├─ weight: 36,864
45
+ ├─ bias: 6
46
+
47
+ 📦 pipe.dit._camera_plucker_patchify_encoder:
48
+ Total params: 35,840
49
+ ├─ weight: 30,720
50
+ ├─ bias: 5,120
51
+
52
+ 📦 pipe.dit.control_patch_embedding:
53
+ Total params: 332,800
54
+ ├─ weight: 327,680
55
+ ├─ bias: 5,120
56
+
57
+ 📦 pipe.dit.domain_embedding:
58
+ Total params: 10,240
59
+ pipe.dit.domain_embedding: 10,240
60
+
61
+ 📦 pipe.track_adapter.conv3d_patchify:
62
+ Total params: 65,568
63
+ ├─ weight: 65,536
64
+ ├─ bias: 32
65
+
66
+ 📦 pipe.track_adapter.mlp:
67
+ Total params: 80,028
68
+ ├─ 0.weight: 30,888
69
+ ├─ 2.weight: 24,336
70
+ ├─ 4.weight: 24,336
71
+ ├─ 0.bias: 156
72
+ ├─ 2.bias: 156
73
+ ├─ 4.bias: 156
74
+
75
+ 📦 pipe.track_adapter.null_token:
76
+ Total params: 156
77
+ pipe.track_adapter.null_token: 156
78
+
79
+ 📦 pipe.track_adapter.proj_down:
80
+ Total params: 5,024
81
+ ├─ weight: 4,992
82
+ ├─ bias: 32
83
+
84
+ 📦 pipe.track_adapter.proj_up:
85
+ Total params: 5,148
86
+ ├─ weight: 4,992
87
+ ├─ bias: 156
88
+
89
+ 📦 pipe.track_block_injector.encoders:
90
+ Total params: 1,080,729,600
91
+ ├─ 0.weight: 27,013,120
92
+ ├─ 1.weight: 27,013,120
93
+ ├─ 10.weight: 27,013,120
94
+ ├─ 11.weight: 27,013,120
95
+ ├─ 12.weight: 27,013,120
96
+ ├─ 13.weight: 27,013,120
97
+ ├─ 14.weight: 27,013,120
98
+ ├─ 15.weight: 27,013,120
99
+ ├─ 16.weight: 27,013,120
100
+ ├─ 17.weight: 27,013,120
101
+ ├─ 18.weight: 27,013,120
102
+ ├─ 19.weight: 27,013,120
103
+ ├─ 2.weight: 27,013,120
104
+ ├─ 20.weight: 27,013,120
105
+ ├─ 21.weight: 27,013,120
106
+ ├─ 22.weight: 27,013,120
107
+ ├─ 23.weight: 27,013,120
108
+ ├─ 24.weight: 27,013,120
109
+ ├─ 25.weight: 27,013,120
110
+ ├─ 26.weight: 27,013,120
111
+ ├─ 27.weight: 27,013,120
112
+ ├─ 28.weight: 27,013,120
113
+ ├─ 29.weight: 27,013,120
114
+ ├─ 3.weight: 27,013,120
115
+ ├─ 30.weight: 27,013,120
116
+ ├─ … (55 more)
117
+
118
+ ────────────────────────────────────────────────────────────────────────────────
119
+ NOTE: FSDP is active; counts below are from the rank-local post-wrap parameter view, not a full global parameter census.
120
+ Total trainable parameters: 6,325,741,170 (6325.74M)
121
+ Total frozen parameters: 10,363,571,200 (10363.57M)
122
+ Percentage trainable (overall): 37.90%
123
+ Percentage trainable (in requested modules: dit): 33.60%
124
+ ================================================================================
125
+
126
+ [torch-profiler] rank0 enabled (wait=3, warmup=1, active=5, repeat=1) -> /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/torch_profiler
127
+ Epoch 1/200: 0%| | 29/7813 [09:35<42:55:34, 19.85s/it, opt_step=2559, loss=0.0606]
128
+ [Track4D-360] First train batch runtime flags: use_gradient_checkpointing=True, use_gradient_checkpointing_offload=False
129
+ [track4d-runtime] rank0 patched DiTBlock.forward saw mode=track block_idx=0
130
+ [Attention] Using: Flash Attention 2 (flash_attn)
131
+ [soft_splat_warp] WARN: tracks grid 160x294 (N=47040) != image grid 288x512 (N=147456); bilinear-resampling tracks_world to image grid. This warning fires once per process.
132
+
133
+ [Training] Reached max_train_steps=2560. Stopping training.
134
+
135
+ [Checkpoint] Saving 593 parameters (total 6,325,741,170 ≈ 6325.74M):
136
+
137
+ 📦 blocks.* — 40 blocks [0..39]:
138
+ Params per block: 131,107,840 × 40 = 5,244,313,600
139
+ Submodules (example from block 0):
140
+ • self_attn: 104,888,320
141
+ • camera_projector: 26,219,520
142
+
143
+ 📦 _camera_patchify_conv.bias:
144
+ Total params: 6
145
+ _camera_patchify_conv.bias: 6
146
+
147
+ 📦 _camera_patchify_conv.weight:
148
+ Total params: 36,864
149
+ _camera_patchify_conv.weight: 36,864
150
+
151
+ 📦 _camera_plucker_patchify_encoder.bias:
152
+ Total params: 5,120
153
+ _camera_plucker_patchify_encoder.bias: 5,120
154
+
155
+ 📦 _camera_plucker_patchify_encoder.weight:
156
+ Total params: 30,720
157
+ _camera_plucker_patchify_encoder.weight: 30,720
158
+
159
+ 📦 control_patch_embedding.bias:
160
+ Total params: 5,120
161
+ control_patch_embedding.bias: 5,120
162
+
163
+ 📦 control_patch_embedding.weight:
164
+ Total params: 327,680
165
+ control_patch_embedding.weight: 327,680
166
+
167
+ 📦 domain_embedding:
168
+ Total params: 10,240
169
+ domain_embedding: 10,240
170
+
171
+ 📦 pipe.dense_target_control_encoder.conv3d_patchify:
172
+ Total params: 10,272
173
+ ├─ weight: 10,240
174
+ ├─ bias: 32
175
+
176
+ 📦 pipe.dense_target_control_encoder.null_token:
177
+ Total params: 156
178
+ pipe.dense_target_control_encoder.null_token: 156
179
+
180
+ 📦 pipe.dense_target_control_encoder.proj_up:
181
+ Total params: 5,148
182
+ ├─ weight: 4,992
183
+ ├─ bias: 156
184
+
185
+ 📦 pipe.dense_target_control_encoder.residual_blocks:
186
+ Total params: 110,720
187
+ ├─ 0.conv1.weight: 27,648
188
+ ├─ 0.conv2.weight: 27,648
189
+ ├─ 1.conv1.weight: 27,648
190
+ ├─ 1.conv2.weight: 27,648
191
+ ├─ 0.conv1.bias: 32
192
+ ├─ 0.conv2.bias: 32
193
+ ├─ 1.conv1.bias: 32
194
+ ├─ 1.conv2.bias: 32
195
+
196
+ 📦 pipe.track_adapter.conv3d_patchify:
197
+ Total params: 65,568
198
+ ├─ weight: 65,536
199
+ ├─ bias: 32
200
+
201
+ 📦 pipe.track_adapter.mlp:
202
+ Total params: 80,028
203
+ ├─ 0.weight: 30,888
204
+ ├─ 2.weight: 24,336
205
+ ├─ 4.weight: 24,336
206
+ ├─ 0.bias: 156
207
+ ├─ 2.bias: 156
208
+ ├─ 4.bias: 156
209
+
210
+ 📦 pipe.track_adapter.null_token:
211
+ Total params: 156
212
+ pipe.track_adapter.null_token: 156
213
+
214
+ 📦 pipe.track_adapter.proj_down:
215
+ Total params: 5,024
216
+ ├─ weight: 4,992
217
+ ├─ bias: 32
218
+
219
+ 📦 pipe.track_adapter.proj_up:
220
+ Total params: 5,148
221
+ ├─ weight: 4,992
222
+ ├─ bias: 156
223
+
224
+ 📦 pipe.track_block_injector.encoders:
225
+ Total params: 1,080,729,600
226
+ ├─ 0.weight: 27,013,120
227
+ ├─ 1.weight: 27,013,120
228
+ ├─ 10.weight: 27,013,120
229
+ ├─ 11.weight: 27,013,120
230
+ ├─ 12.weight: 27,013,120
231
+ ├─ 13.weight: 27,013,120
232
+ ├─ 14.weight: 27,013,120
233
+ ├─ 15.weight: 27,013,120
234
+ … (72 more)
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/files/requirements.txt ADDED
@@ -0,0 +1,208 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ videox-fun==1.0.1
2
+ track4d-360==0.1.0
3
+ func_timeout==4.3.5
4
+ click==8.3.2
5
+ pip==26.0.1
6
+ nvidia-cuda-nvrtc-cu12==12.6.77
7
+ ImageIO==2.37.3
8
+ numpy==1.26.4
9
+ boto3==1.42.91
10
+ albucore==0.0.24
11
+ lpips==0.1.4
12
+ torchsde==0.2.6
13
+ tomesd==0.1.3
14
+ yarl==1.23.0
15
+ typer==0.24.1
16
+ triton==3.5.0
17
+ cycler==0.12.1
18
+ multidict==6.7.1
19
+ OpenEXR==3.4.10
20
+ soxr==1.0.0
21
+ kiwisolver==1.5.0
22
+ simsimd==6.5.16
23
+ parso==0.8.6
24
+ einops==0.8.2
25
+ ftfy==6.3.1
26
+ platformdirs==4.9.6
27
+ groovy==0.1.2
28
+ nvidia-nvtx-cu12==12.6.77
29
+ sentry-sdk==2.58.0
30
+ certifi==2026.2.25
31
+ markdown-it-py==4.0.0
32
+ uvicorn==0.38.0
33
+ nvidia-cudnn-cu12==9.10.2.21
34
+ natsort==8.4.0
35
+ awscrt==0.31.2
36
+ accelerate==1.13.0
37
+ deepspeed==0.18.9
38
+ matplotlib==3.10.8
39
+ mdurl==0.1.2
40
+ regex==2026.4.4
41
+ nvidia-cuda-cupti-cu12==12.6.80
42
+ shellingham==1.5.4
43
+ joblib==1.5.3
44
+ zipp==3.23.1
45
+ ipython_pygments_lexers==1.1.1
46
+ executing==2.2.1
47
+ safetensors==0.7.0
48
+ typing-inspection==0.4.2
49
+ httpcore==1.0.9
50
+ semantic-version==2.10.0
51
+ wandb==0.26.0
52
+ asttokens==3.0.1
53
+ datasets==4.8.4
54
+ decorator==5.2.1
55
+ attrs==26.1.0
56
+ six==1.17.0
57
+ importlib_metadata==9.0.0
58
+ py-cpuinfo==9.0.0
59
+ requests==2.33.1
60
+ xxhash==3.6.0
61
+ pytorch-lightning==2.6.1
62
+ fsspec==2026.2.0
63
+ albumentations==2.0.8
64
+ soundfile==0.13.1
65
+ torchaudio==2.9.0
66
+ torchmetrics==1.9.0
67
+ nvidia-cusolver-cu12==11.7.1.2
68
+ scipy==1.17.1
69
+ jmespath==1.1.0
70
+ threadpoolctl==3.6.0
71
+ plyfile==1.1.2
72
+ peft==0.19.1
73
+ pycparser==3.0
74
+ Markdown==3.10.2
75
+ aiosignal==1.4.0
76
+ pydantic==2.13.2
77
+ decord2==3.3.0
78
+ safehttpx==0.1.7
79
+ scikit-image==0.26.0
80
+ fonttools==4.62.1
81
+ lazy-loader==0.5
82
+ smmap==5.0.3
83
+ starlette==1.0.0
84
+ Werkzeug==3.1.8
85
+ propcache==0.4.1
86
+ tensorboard==2.20.0
87
+ rich==15.0.0
88
+ modelscope==1.36.3
89
+ absl-py==2.4.0
90
+ traitlets==5.14.3
91
+ networkx==3.6.1
92
+ hf-xet==1.4.3
93
+ h5py==3.16.0
94
+ PyYAML==6.0.3
95
+ Jinja2==3.1.6
96
+ nvidia-cusparse-cu12==12.5.4.2
97
+ idna==3.11
98
+ transformers==4.46.2
99
+ httpx==0.28.1
100
+ MarkupSafe==3.0.3
101
+ tensorboard-data-server==0.7.2
102
+ pooch==1.9.0
103
+ torch==2.9.0+cu126
104
+ soupsieve==2.8.3
105
+ numba==0.65.0
106
+ torchdiffeq==0.2.5
107
+ jedi==0.19.2
108
+ charset-normalizer==3.4.7
109
+ annotated-doc==0.0.4
110
+ gradio==6.12.0
111
+ Pygments==2.20.0
112
+ pandas==3.0.2
113
+ stack-data==0.6.3
114
+ contourpy==1.3.3
115
+ tomlkit==0.14.0
116
+ scikit-learn==1.8.0
117
+ librosa==0.11.0
118
+ nvidia-cusparselt-cu12==0.7.1
119
+ pytz==2026.1.post1
120
+ go-with-the-track==0.1.0
121
+ yunchang==0.6.4
122
+ onnxruntime==1.24.4
123
+ ninja==1.13.0
124
+ sentencepiece==0.2.1
125
+ pexpect==4.9.0
126
+ multiprocess==0.70.19
127
+ omegaconf==2.3.0
128
+ h11==0.16.0
129
+ pydub==0.25.1
130
+ nvidia-cufft-cu12==11.3.0.4
131
+ xfuser==0.4.5
132
+ tifffile==2026.3.3
133
+ msgpack==1.1.2
134
+ psutil==7.2.2
135
+ ipython==9.12.0
136
+ hf-gradio==0.4.1
137
+ diffusers==0.33.1
138
+ python-dateutil==2.9.0.post0
139
+ orjson==3.11.8
140
+ huggingface_hub==0.36.2
141
+ nvidia-cuda-runtime-cu12==12.6.77
142
+ beautifulsoup4==4.14.3
143
+ ptyprocess==0.7.0
144
+ protobuf==7.34.1
145
+ GitPython==3.1.46
146
+ sympy==1.14.0
147
+ packaging==26.1
148
+ aiohttp==3.13.5
149
+ nvidia-nvshmem-cu12==3.3.20
150
+ annotated-types==0.7.0
151
+ antlr4-python3-runtime==4.9.3
152
+ gradio_client==2.4.1
153
+ stringzilla==4.6.0
154
+ flatbuffers==25.12.19
155
+ filelock==3.28.0
156
+ imageio-ffmpeg==0.6.0
157
+ fastapi==0.136.0
158
+ nvidia-nvjitlink-cu12==12.6.85
159
+ s3transfer==0.16.0
160
+ python-multipart==0.0.26
161
+ DistVAE==0.0.0b5
162
+ urllib3==2.6.3
163
+ nvidia-nccl-cu12==2.27.5
164
+ av==15.1.0
165
+ pure_eval==0.2.3
166
+ wheel==0.46.3
167
+ dill==0.4.1
168
+ flash_attn==2.8.3
169
+ track4d-360==0.1.0
170
+ opencv-python==4.11.0.86
171
+ anyio==4.13.0
172
+ pytest==9.0.3
173
+ nvidia-cufile-cu12==1.11.1.6
174
+ lightning-utilities==0.15.3
175
+ pluggy==1.6.0
176
+ grpcio==1.80.0
177
+ cffi==2.0.0
178
+ iniconfig==2.3.0
179
+ videox-fun==1.0.1
180
+ matplotlib-inline==0.2.1
181
+ tokenizers==0.20.3
182
+ pydantic_core==2.46.2
183
+ aiohappyeyeballs==2.6.1
184
+ typing_extensions==4.15.0
185
+ timm==1.0.26
186
+ pyarrow==23.0.1
187
+ mpmath==1.3.0
188
+ prompt_toolkit==3.0.52
189
+ mediapy==1.2.6
190
+ llvmlite==0.47.0
191
+ opencv-python-headless==4.11.0.86
192
+ hjson==3.1.0
193
+ tqdm==4.67.3
194
+ torchvision==0.24.0
195
+ pyparsing==3.3.2
196
+ nvidia-cublas-cu12==12.6.4.1
197
+ botocore==1.42.91
198
+ nanobind==2.9.2
199
+ wcwidth==0.6.0
200
+ gitdb==4.0.12
201
+ frozenlist==1.8.0
202
+ brotli==1.2.0
203
+ trampoline==0.1.2
204
+ pillow==12.2.0
205
+ audioread==3.1.0
206
+ lightning==2.6.1
207
+ setuptools==82.0.1
208
+ nvidia-curand-cu12==10.3.7.77
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/files/wandb-metadata.json ADDED
@@ -0,0 +1,235 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-6.4.0-150600.23.73_15.0.14-cray_shasta_c_64k-aarch64-with-glibc2.38",
3
+ "python": "CPython 3.12.13",
4
+ "startedAt": "2026-06-02T06:50:12.043060Z",
5
+ "args": [
6
+ "--model_id_with_origin_paths",
7
+ "Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth",
8
+ "--tokenizer_path",
9
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl",
10
+ "--task",
11
+ "sft:v2v",
12
+ "--model_variant",
13
+ "vt2v",
14
+ "--v2v_concat_mode",
15
+ "width",
16
+ "--height",
17
+ "288",
18
+ "--width",
19
+ "512",
20
+ "--num_frames",
21
+ "49",
22
+ "--track4d360_data_config",
23
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml",
24
+ "--steps_per_epoch",
25
+ "25000",
26
+ "--num_epochs",
27
+ "200",
28
+ "--learning_rate",
29
+ "2e-5",
30
+ "--gradient_accumulation_steps",
31
+ "1",
32
+ "--save_steps",
33
+ "500",
34
+ "--remove_prefix_in_ckpt",
35
+ "pipe.dit.",
36
+ "--trainable_models",
37
+ "dit.domain_embedding,dit.control_patch_embedding",
38
+ "--dit_finetune_modules",
39
+ "self_attn",
40
+ "--track_injection_mode",
41
+ "per_block",
42
+ "--track_injection_block_mode",
43
+ "concat_project",
44
+ "--track_config",
45
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth_noconv3dres.yaml",
46
+ "--train_sigma_shift",
47
+ "5.0",
48
+ "--use_query_frame_impulse_condition",
49
+ "--use_plucker_camera_control",
50
+ "--enable_v2v_plucker_camera_control",
51
+ "--use_plucker_raymap_tokens",
52
+ "--use_perlayer_camera_control",
53
+ "--camera_injection_mode",
54
+ "projector",
55
+ "--camera_plucker_encoder_mode",
56
+ "patchify",
57
+ "--use_dense_branch",
58
+ "--dense_in_channels",
59
+ "5",
60
+ "--dense_proj_dim",
61
+ "32",
62
+ "--dense_num_residual_blocks",
63
+ "2",
64
+ "--dense_alpha_track",
65
+ "1.0",
66
+ "--dense_conditioning_noise_max_level",
67
+ "0.15",
68
+ "--source_mask_boundary_depth_dropout_prob",
69
+ "0.5",
70
+ "--source_mask_boundary_kernel_size",
71
+ "5",
72
+ "--dense_warp_mode",
73
+ "hybrid_softsplat",
74
+ "--dense_warp_motion_threshold",
75
+ "0.005",
76
+ "--cond_drop_geo_prob",
77
+ "0.1",
78
+ "--cond_drop_track_prob",
79
+ "0.1",
80
+ "--cond_drop_dense_prob",
81
+ "0.0",
82
+ "--initial_step",
83
+ "2530",
84
+ "--output_path",
85
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout",
86
+ "--dataset_num_workers",
87
+ "7",
88
+ "--dataloader_prefetch_factor",
89
+ "2",
90
+ "--drop_text_prob",
91
+ "0.1",
92
+ "--debug_latents_every",
93
+ "500",
94
+ "--allow_existing_checkpoints",
95
+ "--wandb_project",
96
+ "track4d-360",
97
+ "--wandb_name",
98
+ "ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp",
99
+ "--wandb_tags",
100
+ "ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug",
101
+ "--use_wandb",
102
+ "--vxf_init_checkpoint",
103
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors",
104
+ "--use_gradient_checkpointing",
105
+ "--resume_checkpoint",
106
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors",
107
+ "--max_train_steps",
108
+ "2560",
109
+ "--torch_profile_active_steps",
110
+ "5",
111
+ "--torch_profile_wait_steps",
112
+ "3",
113
+ "--torch_profile_warmup_steps",
114
+ "1",
115
+ "--torch_profile_repeat",
116
+ "1",
117
+ "--no_torch_profile_record_shapes",
118
+ "--no_torch_profile_memory"
119
+ ],
120
+ "program": "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/scripts/wan2.1_fun/train_track4d360.py",
121
+ "codePath": "scripts/wan2.1_fun/train_track4d360.py",
122
+ "codePathLocal": "scripts/wan2.1_fun/train_track4d360.py",
123
+ "git": {
124
+ "remote": "git@github.com:NIRVANALAN/track4d-360.git",
125
+ "commit": "b6681ed2127310d14bc5b93ae5f1d20df8b9e907"
126
+ },
127
+ "email": "yushi.lan@outlook.com",
128
+ "root": "/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout",
129
+ "host": "nid010301",
130
+ "executable": "/projects/public/u6ew/yslan_public/yslan_project/miniforge3/envs/th29py312/bin/python3.12",
131
+ "cpu_count": 288,
132
+ "cpu_count_logical": 288,
133
+ "gpu": "NVIDIA GH200 120GB",
134
+ "gpu_count": 4,
135
+ "disk": {
136
+ "/": {
137
+ "total": "255795724288",
138
+ "used": "546373632"
139
+ }
140
+ },
141
+ "memory": {
142
+ "total": "919613407232"
143
+ },
144
+ "gpu_nvidia": [
145
+ {
146
+ "name": "NVIDIA GH200 120GB",
147
+ "memoryTotal": "102625181696",
148
+ "cudaCores": 16896,
149
+ "architecture": "Hopper",
150
+ "uuid": "GPU-fda2ad82-f650-d1df-fbce-bab2bc8dd421"
151
+ },
152
+ {
153
+ "name": "NVIDIA GH200 120GB",
154
+ "memoryTotal": "102625181696",
155
+ "cudaCores": 16896,
156
+ "architecture": "Hopper",
157
+ "uuid": "GPU-f29ba6f4-4da2-cc4b-37dc-d260fa79a222"
158
+ },
159
+ {
160
+ "name": "NVIDIA GH200 120GB",
161
+ "memoryTotal": "102625181696",
162
+ "cudaCores": 16896,
163
+ "architecture": "Hopper",
164
+ "uuid": "GPU-38da6d83-51b2-5586-2248-c681407f2a51"
165
+ },
166
+ {
167
+ "name": "NVIDIA GH200 120GB",
168
+ "memoryTotal": "102625181696",
169
+ "cudaCores": 16896,
170
+ "architecture": "Hopper",
171
+ "uuid": "GPU-aa24fa01-5dca-c359-28f7-7adf38ef534a"
172
+ }
173
+ ],
174
+ "cudaVersion": "12.7",
175
+ "slurm": {
176
+ "cluster_name": "gracehopper",
177
+ "conf": "/var/spool/slurmd/conf-cache/slurm.conf",
178
+ "cpu_bind": "quiet,mask_cpu:0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF",
179
+ "cpu_bind_list": "0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF",
180
+ "cpu_bind_type": "mask_cpu:",
181
+ "cpu_bind_verbose": "quiet",
182
+ "cpus_on_node": "288",
183
+ "distribution": "cyclic",
184
+ "gpus": "32",
185
+ "gpus_on_node": "4",
186
+ "gpus_per_node": "4",
187
+ "gtids": "0",
188
+ "job_account": "brics.u6ew",
189
+ "job_cpus_per_node": "288(x8)",
190
+ "job_end_time": "1780389885",
191
+ "job_gid": "1483804847",
192
+ "job_gpus": "0,1,2,3",
193
+ "job_id": "4982942",
194
+ "job_name": "14b_pair1x49_profile_ablate_conv3dres0_fsdp2",
195
+ "job_nodelist": "nid[010301-010304,010306,010308-010309,010321]",
196
+ "job_num_nodes": "8",
197
+ "job_partition": "workq",
198
+ "job_qos": "normal",
199
+ "job_start_time": "1780382685",
200
+ "job_uid": "1483804847",
201
+ "job_user": "yushi.u6ew",
202
+ "jobid": "4982942",
203
+ "launch_node_ipaddr": "10.242.5.140",
204
+ "localid": "0",
205
+ "mpi_type": "pmi2",
206
+ "nnodes": "8",
207
+ "nodeid": "0",
208
+ "nodelist": "nid[010301-010304,010306,010308-010309,010321]",
209
+ "nprocs": "8",
210
+ "ntasks": "8",
211
+ "ntasks_per_node": "1",
212
+ "oom_kill_step": "0",
213
+ "prio_process": "0",
214
+ "procid": "0",
215
+ "srun_comm_host": "10.242.5.140",
216
+ "srun_comm_port": "45579",
217
+ "step_gpus": "0,1,2,3",
218
+ "step_id": "0",
219
+ "step_launcher_port": "45579",
220
+ "step_nodelist": "nid[010301-010304,010306,010308-010309,010321]",
221
+ "step_num_nodes": "8",
222
+ "step_num_tasks": "8",
223
+ "step_resv_ports": "26618-26619",
224
+ "step_tasks_per_node": "1(x8)",
225
+ "stepid": "0",
226
+ "submit_dir": "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360",
227
+ "submit_host": "login40",
228
+ "task_pid": "152888",
229
+ "tasks_per_node": "1(x8)",
230
+ "topology_addr": "global.group4.nid010301",
231
+ "topology_addr_pattern": "switch.switch.node",
232
+ "umask": "0022"
233
+ },
234
+ "writerId": "1m9b4etgqyfvjqvydz51tku5m7qgycq3"
235
+ }
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/files/wandb-summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"train/track/injector/block_19/weight_norm":71.72638702392578,"train/track/injector/block_12/grad_norm":0.005037492141127586,"train/timing/unit/WanVideoUnit_Vista4DStreams_s":4.249280391377397e-05,"train/track/injector/block_16/weight_norm":71.75887298583984,"train/track/injector/block_14/weight_norm":71.7538833618164,"train/track/injector/block_06/weight_norm":71.78463745117188,"train/timing/unit/WanVideoUnit_TrackPointCondition_s":0.02038385346531868,"train/track/injector/block_03/grad_norm":0.008266760036349297,"train/track/injector/grad_norm_max":0.025773145258426666,"train/track/injector/block_38/weight_norm":71.60935974121094,"train/timing/data_wait_s":0.0011231391690671444,"train/track/injector/block_07/grad_norm":0.006979101803153753,"train/timing/unit/WanVideoUnit_CameraControl_NVS_s":0.003995758481323719,"train/track/injector/block_09/grad_norm":0.006004222668707371,"train/timing/unit/WanVideoUnit_ShapeChecker_s":4.563275069813244e-05,"train/timing/dit_loss_fn_s":2.957465648651123,"train/loss":0.060648560523986816,"train/track/injector/block_27/weight_norm":71.7016830444336,"train/track/injector/block_31/weight_norm":71.70208740234375,"train/hbm/reserved_gb":39.54953384399414,"train/track/injector/block_32/weight_norm":71.66666412353516,"train/track/injector/block_04/grad_norm":0.007650282233953476,"train/track/injector/block_23/weight_norm":71.71662902832031,"train/track/injector/block_25/weight_norm":71.7304916381836,"train/track/injector/block_30/grad_norm":0.002275619190186262,"train/timing/unit/WanVideoUnit_PromptEmbedder_s":0.00026011947193183005,"train/track/injector/block_22/weight_norm":71.71588897705078,"train/track/injector/block_00/grad_norm":0.025773145258426666,"train/track/injector/block_16/grad_norm":0.005171981640160084,"train/track/injector/block_05/weight_norm":71.7911376953125,"train/track/injector/block_13/weight_norm":71.7365493774414,"train/track/injector/block_37/grad_norm":0.0021059231366962194,"train/track/injector/block_05/grad_norm":0.007884364575147629,"train/track/injector/block_08/weight_norm":71.7770767211914,"train/track/injector/block_15/weight_norm":71.72616577148438,"train/timing/reduction_s":0.10025622975081205,"train/track/injector/block_00/weight_norm":71.61991882324219,"train/cond_dropout/drop_dense":0.09375,"train/track/injector/block_22/grad_norm":0.004614893347024918,"train/track/injector/block_02/grad_norm":0.008216189220547676,"train/timing/step_total_s":13.46668770769611,"train/track/injector/block_13/grad_norm":0.005134806968271732,"train/track/injector/block_12/weight_norm":71.75670623779297,"train/track/injector/block_07/weight_norm":71.78170013427734,"train/track/injector/block_14/grad_norm":0.005121688358485699,"train/track/injector/block_06/grad_norm":0.007252744399011135,"_wandb":{"runtime":598},"train/track/injector/block_37/weight_norm":71.64714813232422,"train/timing/unit/WanVideoUnit_TargetVideoEncoder_s":0.8007816672325134,"train/track/injector/block_01/grad_norm":0.008102822117507458,"train/track/injector/block_21/grad_norm":0.003961995244026184,"train/track/injector/block_30/weight_norm":71.7083511352539,"train/track/injector/block_20/grad_norm":0.0034632328897714615,"train/track/injector/block_15/grad_norm":0.004622212611138821,"train/track/injector/weight_norm_mean":71.7210922241211,"train/track/injector/block_39/grad_norm":0.005907048471271992,"train/track/injector/block_36/weight_norm":71.649658203125,"train/hbm/allocated_gb":3.435471534729004,"train/track/adapter/grad_norm":0.004346003755927086,"train/track/injector/block_32/grad_norm":0.0021721147932112217,"train/track/injector/block_23/grad_norm":0.0035610890481621027,"train/track/injector/block_28/weight_norm":71.74327850341797,"train/timing/unit/WanVideoUnit_TeaCache_s":1.821445403038524e-05,"train/track/injector/block_04/weight_norm":71.78107452392578,"train/track/injector/block_09/weight_norm":71.78056335449219,"train/hbm/post_t5_allocated_gb":28.869243621826172,"train/track/injector/block_31/grad_norm":0.0020449731964617968,"train/track/injector/block_38/grad_norm":0.002583117689937353,"train/timing/unit/WanVideoUnit_DenseTargetControl_s":0.2740522027015686,"train/track/injector/block_03/weight_norm":71.80154418945312,"train/timing/unit/WanVideoUnit_V2VConcatenate_s":0.00011662059841910377,"train/track/injector/block_33/grad_norm":0.002395837800577283,"train/track/injector/block_24/weight_norm":71.72344207763672,"train/t5/on_cuda":-1,"train/timing/unit/WanVideoUnit_LatentNoiseAugmentation_s":4.298090061638504e-05,"train/track/injector/weight_norm_max":71.80154418945312,"train/track/injector/block_35/weight_norm":71.68324279785156,"train/track/injector/block_21/weight_norm":71.71271514892578,"train/cond_dropout/drop_tracks":0.15625,"_runtime":598,"train/timing/optimizer_s":0.031879177782684565,"train/track/injector/grad_norm_mean":0.005017243325710297,"train/track/injector/block_18/grad_norm":0.004682852420955896,"train/track/injector/block_11/grad_norm":0.005184764973819256,"train/timing/unit/WanVideoUnit_WarpedAppearance_s":4.666278255172074e-05,"train/track/injector/block_36/grad_norm":0.0020740143954753876,"train/epoch":0,"train/timing/unit/WanVideoUnit_NoiseInitializer_s":0.00017827811825554818,"train/track/injector/block_35/grad_norm":0.0019575022161006927,"train/track/injector/block_34/grad_norm":0.00215382338501513,"train/timing/unit/WanVideoUnit_InputVideoEmbedder_XYZCondition_s":1.6093270778656006,"train/timing/unit/WanVideoUnit_CfgMerger_s":1.1475616702227853e-05,"train/timing/microsteps":1,"train/track/injector/block_25/grad_norm":0.0028981242794543505,"train/track/injector/block_29/weight_norm":71.7083969116211,"train/track/injector/block_34/weight_norm":71.64429473876953,"train/track/injector/block_26/weight_norm":71.73137664794922,"_step":2559,"train/track/injector/block_39/weight_norm":71.61332702636719,"train/track/injector/block_17/grad_norm":0.004457436036318541,"train/track/injector/block_33/weight_norm":71.65998077392578,"train/track/injector/block_17/weight_norm":71.72579956054688,"train/track/injector/block_26/grad_norm":0.002922400366514921,"train/track/injector/block_10/grad_norm":0.0058774868957698345,"train/track/injector/block_20/weight_norm":71.71338653564453,"train/hbm/post_t5_reserved_gb":28.94489097595215,"train/track/injector/block_19/grad_norm":0.00396856339648366,"train/track/injector/block_11/weight_norm":71.75855255126953,"train/timing/other_s":1.90594895882532,"_timestamp":1.7803835989144747e+09,"train/track/injector/block_27/grad_norm":0.0028194861952215433,"train/track/injector/block_24/grad_norm":0.0033174175769090652,"train/timing/unit/WanVideoUnit_UnifiedSequenceParallel_s":2.4413131541223265e-05,"train/track/injector/block_29/grad_norm":0.0028786961920559406,"train/track/injector/block_18/weight_norm":71.71785736083984,"train/timing/unit/WanVideoUnit_ModeDropoutAugmentation_s":2.5872304831864312e-05,"train/timing/backward_s":5.718927487265319,"train/track/injector/block_08/grad_norm":0.006412863731384277,"train/track/adapter/weight_norm":16.3490047454834,"train/timing/forward_s":5.709675854071975,"train/track/injector/block_01/weight_norm":71.73580932617188,"train/track/injector/block_02/weight_norm":71.77042388916016,"train/track/injector/block_10/weight_norm":71.77764129638672,"train/track/injector/block_28/grad_norm":0.0027806407306343317}
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/logs/debug-internal.log ADDED
@@ -0,0 +1,97 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-06-02T06:50:13.570807992Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-06-02T06:50:13.573515504Z","level":"INFO","msg":"stream: starting","core version":"0.26.0"}
3
+ {"time":"2026-06-02T06:50:14.064306628Z","level":"INFO","msg":"stream: created new stream","id":"ixxjov67"}
4
+ {"time":"2026-06-02T06:50:14.064412769Z","level":"INFO","msg":"handler: started"}
5
+ {"time":"2026-06-02T06:50:14.065276714Z","level":"INFO","msg":"stream: started"}
6
+ {"time":"2026-06-02T06:50:14.065282794Z","level":"INFO","msg":"writer: started","stream_id":"ixxjov67"}
7
+ {"time":"2026-06-02T06:50:14.06566832Z","level":"INFO","msg":"sender: started"}
8
+ {"time":"2026-06-02T06:50:15.631914228Z","level":"INFO","msg":"filestream: sending request","total_files":1,"console_offset":0,"console_lines":1}
9
+ {"time":"2026-06-02T06:50:16.106065701Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
10
+ {"time":"2026-06-02T06:50:23.169926895Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":44}
11
+ {"time":"2026-06-02T06:50:23.18417797Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":175}
12
+ {"time":"2026-06-02T06:50:30.632687583Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":0,"events_lines":2,"console_offset":1,"console_lines":126,"uploaded_len":2}
13
+ {"time":"2026-06-02T06:50:31.057643889Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
14
+ {"time":"2026-06-02T06:50:45.633956602Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":2,"events_lines":2}
15
+ {"time":"2026-06-02T06:50:45.995092564Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
16
+ {"time":"2026-06-02T06:51:00.632732136Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":4,"events_lines":2}
17
+ {"time":"2026-06-02T06:51:01.013164986Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
18
+ {"time":"2026-06-02T06:51:15.632693827Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":6,"events_lines":2}
19
+ {"time":"2026-06-02T06:51:15.973083783Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
20
+ {"time":"2026-06-02T06:51:30.632614271Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":8,"events_lines":2}
21
+ {"time":"2026-06-02T06:51:30.902457359Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
22
+ {"time":"2026-06-02T06:51:45.632638154Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":10,"events_lines":2,"console_offset":127,"console_lines":3}
23
+ {"time":"2026-06-02T06:51:46.012081416Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
24
+ {"time":"2026-06-02T06:52:00.632782948Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":0,"history_lines":1,"events_offset":12,"events_lines":2,"console_offset":126,"console_lines":1}
25
+ {"time":"2026-06-02T06:52:01.245405887Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
26
+ {"time":"2026-06-02T06:52:15.633245252Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":1,"history_lines":1,"events_offset":14,"events_lines":2,"console_offset":126,"console_lines":1}
27
+ {"time":"2026-06-02T06:52:16.031118944Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
28
+ {"time":"2026-06-02T06:52:30.632460243Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":2,"history_lines":2,"events_offset":16,"events_lines":2,"console_offset":126,"console_lines":1}
29
+ {"time":"2026-06-02T06:52:31.064836975Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
30
+ {"time":"2026-06-02T06:52:45.632586315Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":4,"history_lines":1,"events_offset":18,"events_lines":2,"console_offset":126,"console_lines":1}
31
+ {"time":"2026-06-02T06:52:45.920833256Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
32
+ {"time":"2026-06-02T06:53:00.632560879Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":5,"history_lines":1,"events_offset":20,"events_lines":2,"console_offset":126,"console_lines":1}
33
+ {"time":"2026-06-02T06:53:01.114903023Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
34
+ {"time":"2026-06-02T06:53:15.633256291Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":6,"history_lines":1,"events_offset":22,"events_lines":2,"console_offset":126,"console_lines":1}
35
+ {"time":"2026-06-02T06:53:16.048155401Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
36
+ {"time":"2026-06-02T06:53:30.633585048Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":7,"history_lines":1,"events_offset":24,"events_lines":2,"console_offset":126,"console_lines":1}
37
+ {"time":"2026-06-02T06:53:31.116840996Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
38
+ {"time":"2026-06-02T06:53:45.632811032Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":8,"history_lines":1,"events_offset":26,"events_lines":2,"console_offset":126,"console_lines":1}
39
+ {"time":"2026-06-02T06:53:45.995726219Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
40
+ {"time":"2026-06-02T06:54:00.632163343Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":28,"events_lines":2,"console_offset":130,"console_lines":1}
41
+ {"time":"2026-06-02T06:54:01.04412747Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
42
+ {"time":"2026-06-02T06:54:15.632919183Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":30,"events_lines":2}
43
+ {"time":"2026-06-02T06:54:16.033687178Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
44
+ {"time":"2026-06-02T06:54:30.632581426Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":32,"events_lines":2}
45
+ {"time":"2026-06-02T06:54:31.115164341Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
46
+ {"time":"2026-06-02T06:54:45.631988187Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":34,"events_lines":2}
47
+ {"time":"2026-06-02T06:54:46.16067707Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
48
+ {"time":"2026-06-02T06:55:00.632478441Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":36,"events_lines":2}
49
+ {"time":"2026-06-02T06:55:01.137467668Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
50
+ {"time":"2026-06-02T06:55:15.634021135Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":38,"events_lines":2}
51
+ {"time":"2026-06-02T06:55:16.528050934Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
52
+ {"time":"2026-06-02T06:55:30.633071312Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":9,"history_lines":1,"events_offset":40,"events_lines":2,"console_offset":126,"console_lines":1}
53
+ {"time":"2026-06-02T06:55:30.990818571Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
54
+ {"time":"2026-06-02T06:55:45.632948623Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":10,"history_lines":1,"events_offset":42,"events_lines":2,"console_offset":126,"console_lines":1}
55
+ {"time":"2026-06-02T06:55:45.947339192Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
56
+ {"time":"2026-06-02T06:56:00.63331661Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":11,"history_lines":1,"events_offset":44,"events_lines":2,"console_offset":126,"console_lines":1}
57
+ {"time":"2026-06-02T06:56:01.090740183Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
58
+ {"time":"2026-06-02T06:56:15.633357213Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":12,"history_lines":1,"events_offset":46,"events_lines":2,"console_offset":126,"console_lines":1}
59
+ {"time":"2026-06-02T06:56:16.100888881Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
60
+ {"time":"2026-06-02T06:56:30.633508896Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":13,"history_lines":2,"events_offset":48,"events_lines":2,"console_offset":126,"console_lines":1}
61
+ {"time":"2026-06-02T06:56:31.248935786Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
62
+ {"time":"2026-06-02T06:56:45.633656982Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":15,"history_lines":1,"events_offset":50,"events_lines":2,"console_offset":126,"console_lines":1}
63
+ {"time":"2026-06-02T06:56:46.023339526Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
64
+ {"time":"2026-06-02T06:57:00.633242778Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":16,"history_lines":1,"events_offset":52,"events_lines":2,"console_offset":126,"console_lines":1}
65
+ {"time":"2026-06-02T06:57:01.000390992Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
66
+ {"time":"2026-06-02T06:57:15.632535397Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":17,"history_lines":2,"events_offset":54,"events_lines":2,"console_offset":126,"console_lines":1}
67
+ {"time":"2026-06-02T06:57:16.015127404Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
68
+ {"time":"2026-06-02T06:57:30.633130173Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":19,"history_lines":1,"events_offset":56,"events_lines":2,"console_offset":126,"console_lines":1}
69
+ {"time":"2026-06-02T06:57:30.99356342Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
70
+ {"time":"2026-06-02T06:57:45.633329584Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":20,"history_lines":1,"events_offset":58,"events_lines":2,"console_offset":126,"console_lines":1}
71
+ {"time":"2026-06-02T06:57:46.07541001Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
72
+ {"time":"2026-06-02T06:58:00.632386559Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":21,"history_lines":1,"events_offset":60,"events_lines":2,"console_offset":126,"console_lines":1}
73
+ {"time":"2026-06-02T06:58:01.18224948Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
74
+ {"time":"2026-06-02T06:58:15.632888861Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":22,"history_lines":2,"events_offset":62,"events_lines":2,"console_offset":126,"console_lines":1}
75
+ {"time":"2026-06-02T06:58:16.030413718Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
76
+ {"time":"2026-06-02T06:58:30.633023024Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":24,"history_lines":1,"events_offset":64,"events_lines":2,"console_offset":126,"console_lines":1}
77
+ {"time":"2026-06-02T06:58:31.068636213Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
78
+ {"time":"2026-06-02T06:58:45.633331848Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":25,"history_lines":1,"events_offset":66,"events_lines":2,"console_offset":126,"console_lines":1}
79
+ {"time":"2026-06-02T06:58:46.039435816Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
80
+ {"time":"2026-06-02T06:59:00.632360803Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":26,"history_lines":1,"events_offset":68,"events_lines":2,"console_offset":126,"console_lines":1}
81
+ {"time":"2026-06-02T06:59:01.039092307Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
82
+ {"time":"2026-06-02T06:59:15.632705468Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":70,"events_lines":2}
83
+ {"time":"2026-06-02T06:59:16.085584263Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
84
+ {"time":"2026-06-02T06:59:30.632579618Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":72,"events_lines":2}
85
+ {"time":"2026-06-02T06:59:31.365655224Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
86
+ {"time":"2026-06-02T06:59:45.633048671Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":27,"history_lines":2,"events_offset":74,"events_lines":2,"console_offset":126,"console_lines":1}
87
+ {"time":"2026-06-02T06:59:46.003907075Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
88
+ {"time":"2026-06-02T07:00:00.633135873Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":29,"history_lines":1,"events_offset":76,"events_lines":2,"console_offset":126,"console_lines":1}
89
+ {"time":"2026-06-02T07:00:01.363370609Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
90
+ {"time":"2026-06-02T07:00:13.67931977Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
91
+ {"time":"2026-06-02T07:00:13.679909675Z","level":"INFO","msg":"filestream: sending request","total_files":2,"console_offset":131,"console_lines":103,"uploaded_len":3,"complete":true,"exit_code":0}
92
+ {"time":"2026-06-02T07:00:14.181094862Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
93
+ {"time":"2026-06-02T07:00:14.18215301Z","level":"INFO","msg":"handler: operation stats","stats":{}}
94
+ {"time":"2026-06-02T07:00:14.188249365Z","level":"INFO","msg":"stream: closing"}
95
+ {"time":"2026-06-02T07:00:14.188298388Z","level":"INFO","msg":"handler: closed"}
96
+ {"time":"2026-06-02T07:00:14.188708233Z","level":"INFO","msg":"sender: closed"}
97
+ {"time":"2026-06-02T07:00:14.188737768Z","level":"INFO","msg":"stream: closed"}
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/logs/debug.log ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-06-02 06:50:12,052 INFO MainThread:152969 [wandb_setup.py:_flush():81] Current SDK version is 0.26.0
2
+ 2026-06-02 06:50:12,053 INFO MainThread:152969 [wandb_setup.py:_flush():81] Configure stats pid to 152969
3
+ 2026-06-02 06:50:12,053 INFO MainThread:152969 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-06-02 06:50:12,053 INFO MainThread:152969 [wandb_init.py:setup_run_log_directory():721] Logging user logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/logs/debug.log
5
+ 2026-06-02 06:50:12,054 INFO MainThread:152969 [wandb_init.py:setup_run_log_directory():722] Logging internal logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/logs/debug-internal.log
6
+ 2026-06-02 06:50:12,054 INFO MainThread:152969 [wandb_init.py:init():848] calling init triggers
7
+ 2026-06-02 06:50:12,060 INFO MainThread:152969 [wandb_init.py:init():853] wandb.init called with sweep_config: {}
8
+ config: {'learning_rate': 2e-05, 'weight_decay': 0.01, 'num_epochs': 200, 'save_steps': 500, 'num_workers': 7, 'prefetch_factor': 2, 'dataset_base_path': None, 'dataset_metadata_path': None, 'dataset_repeat': 1, 'dataset_num_workers': 7, 'dataloader_prefetch_factor': 2, 'data_file_keys': 'image,video', 'pointmap_stats_path': None, 'model_paths': None, 'model_id_with_origin_paths': 'Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth', 'model_variant': 'vt2v', 'extra_inputs': None, 'track_xyz_normalization': 'minmax', 'track_xyz_quantile_low': 0.02, 'track_xyz_quantile_high': 0.98, 'predict_track_residual': False, 'residual_target_xyz_source': 'clean', 'residual_zero_center_normalization': False, 'residual_linear_normalization': False, 'residual_log_scale_factor': 3.0, 'stream3r_common_scale_stat': 'mean', 'explicit_motion_loss_weight': 0.0, 'explicit_motion_loss_snr_weight': True, 'explicit_motion_loss_snr_sigmoid': True, 'explicit_motion_loss_snr_sigmoid_a': 1.0, 'explicit_motion_loss_snr_sigmoid_b': -5.0, 'explicit_motion_loss_decode_checkpoint': False, 'explicit_motion_loss_decode_chunk_size': 0, 'normalize_track_with_xyz_cond_stats': False, 'prenormalized_track_video_signed_log': False, 'use_xyz_video_condition': False, 'xyz_video_condition_scale': 1.0, 'xyz_video_condition_aug_prob': 0.0, 'xyz_video_condition_noise_std': 0.0, 'xyz_video_condition_z_noise_std': 0.0, 'xyz_video_condition_frame_scale_std': 0.0, 'xyz_video_condition_dropout_prob': 0.0, 'xyz_video_condition_mode': 'xyz', 'xyz_video_condition_cfg_dropout_prob': 0.0, 'camera_condition_cfg_dropout_prob': 0.0, 'fp8_models': None, 'offload_models': None, 'v2v_concat_mode': 'width', 'v2v_channel_patch_embed_mode': 'add', 'trainable_models': 'dit.domain_embedding,dit.control_patch_embedding', 'find_unused_parameters': False, 'task': 'sft:v2v', 'drop_text_prob': 0.1, 'drop_video_prob': 0.0, 'drop_t2v_prob': None, 'drop_i2v_prob': None, 'input_video_noise_strength': 0.0, 'input_video_noise_random_max': 500, 'output_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout', 'remove_prefix_in_ckpt': 'pipe.dit.', 'save_every_n_epochs': 2, 'max_checkpoints_to_keep': 5, 'allow_existing_checkpoints': True, 'initial_step': 2530, 'max_train_steps': 2560, 'lora_base_model': None, 'lora_target_modules': 'q,k,v,o,ffn.0,ffn.2', 'lora_rank': 32, 'lora_alpha': None, 'lora_checkpoint': None, 'preset_lora_path': None, 'preset_lora_model': None, 'use_gradient_checkpointing': True, 'use_gradient_checkpointing_offload': False, 'gradient_accumulation_steps': 1, 'use_wandb': True, 'wandb_project': 'track4d-360', 'wandb_name': 'ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp', 'wandb_entity': None, 'wandb_tags': ['ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug'], 'wandb_notes': None, 'wandb_log_gradients': False, 'wandb_log_parameters': False, 'wandb_log_freq': 100, 'wandb_watch_model': False, 'debug_latents': False, 'debug_latents_every': 500, 'debug_latents_fps': 16, 'debug_latents_quality': 5, 'debug_latents_max_frames': None, 'debug_latents_outdir': 'debug_latents', 'debug_latents_vae_tiled': True, 'debug_latents_vae_tile_size': (30, 52), 'debug_latents_vae_tile_stride': (15, 26), 'debug_missing_grads': False, 'torch_distributed_debug': None, 'use_query_frame_impulse_condition': True, 'height': 288, 'width': 512, 'max_pixels': 1048576, 'num_frames': 49, 'tokenizer_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl', 'dit_finetune_modules': 'self_attn', 'track_injection_mode': 'per_block', 'track_injection_block_mode': 'concat_project', 'train_sigma_shift': 5.0, 'dummy': False, 'steps_per_epoch': 25000, 'initialize_model_on_cpu': False, 'resume_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors', 'gs_head_init_checkpoint': None, 'joint_track_prediction': False, 'joint_track_depth_prediction': False, 'joint_depth_prediction': False, 'target_depth_log_cap': 14.0, 'target_depth_norm_type': 'log_depth', 'target_depth_max_disparity': 5.0, 'vxf_init_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors', 'channel_concat_stage': None, 'use_residual_tracks': True, 'freeze_lora_after_load': False, 'skip_v2v_required_trainables': False, 'gs_only_training': False, 'freeze_track_block_injector': False, 'freeze_track4d_condition_modules': False, 'gs_head_style': 'flashworld', 'use_gs_output': False, 'gs_zero_features': False, 'gs_feat_dim': 1024, 'gs_loss_weight': 1.0, 'gs_mse_loss_weight': 1.0, 'gs_lpips_loss_weight': 0.5, 'gs_depth_loss_weight': 0.05, 'gs_abs_depth_loss_weight': 0.0, 'gs_opacity_loss_weight': 0.1, 'gs_opacity_prune_keep_frac': 1.0, 'gs_rgb_loss_weight': None, 'gs_sh_degree': 2, 'gs_num_points_per_pixel': 2, 'no_gs_latent_detach': False, 'use_gs_feedback_training': False, 'gs_scale_cap': None, 'gs_render_patch_size': None, 'gs_legacy_union_render': False, 'gs_lyra_depth_param': 'exp', 'gs_lyra_log_depth_shift': 0.0, 'gs_lyra_dnear': 0.1, 'gs_lyra_dfar': 10.0, 'gs_replay_cache_size': 0, 'gs_dynamic_mode': 'static', 'gs_track_supervision_mode': 'pointmap', 'gs_pos_loss_weight': 1.0, 'gs_dyn_depth_loss_weight': 0.5, 'gs_opa_tv_weight': 0.01, 'gs_xyz_tv_weight': 0.01, 'gs_predict_rotation': False, 'gs_disable_delta_opacity': False, 'gs_direct_pos_loss_weight': 0.0, 'gs_track_canonical_loss_weight': 0.0, 'gs_track_delta_loss_weight': 0.0, 'gs_track_final_pos_loss_weight': 0.0, 'gs_dynamic_max_t_video': 64, 'gs_assoc_track_2d_loss_weight': 0.0, 'gs_track_accel_loss_weight': 0.0, 'gs_track_loss_quantile': 0.98, 'gs_track_canonical_detach': False, 'gs_track_joint_render': False, 'cond_drop_track_prob': 0.1, 'cond_drop_dense_prob': 0.0, 'cond_drop_geo_prob': 0.1, 'use_track_null_token_fallback': False, 'use_dense_null_token_fallback': False, 'use_plucker_camera_control': True, 'enable_v2v_plucker_camera_control': True, 'train_plucker_camera_control': False, 'use_plucker_raymap_tokens': True, 'plucker_use_ray_origin': False, 'use_perlayer_camera_control': True, 'camera_injection_mode': 'projector', 'camera_plucker_encoder_mode': 'patchify', 'warped_condition_mode': 'off', 'warped_appearance_fusion': 'concat_proj', 'warped_geom_only_dense': True, 'multi_frame_warped_appearance': False, 'multi_frame_motion_threshold': 0.01, 'fixed_step_frac': None, 'fixed_num_inference_steps': 50, 'fixed_sigma': None, 'skip_diffusion_loss': False, 'use_dense_branch': True, 'dense_in_channels': 5, 'dense_proj_dim': 32, 'dense_num_residual_blocks': 2, 'dense_alpha_track': 1.0, 'use_depth_null_token': False, 'dense_conditioning_noise_max_level': 0.15, 'source_mask_boundary_depth_dropout_prob': 0.5, 'source_mask_boundary_kernel_size': 5, 'dense_warp_mode': 'hybrid_softsplat', 'dense_warp_motion_threshold': 0.005, 'dense_hybrid_disable_dynamic_masks': False, 'use_vista4d_latent_streams': False, 'use_gen3c_warp_latent_channels': False, 'disable_track_modules': False, 'dense_only_condition_modules': False, 'torch_profile_active_steps': 5, 'torch_profile_wait_steps': 3, 'torch_profile_warmup_steps': 1, 'torch_profile_repeat': 1, 'torch_profile_dir': None, 'torch_profile_record_shapes': False, 'torch_profile_memory': False, 'torch_profile_with_stack': False, 'track_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth_noconv3dres.yaml', 'track4d360_data_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml', 'syn4d_root': '/scratch/shared/beegfs/zeren/Syn4D', 'scene': None, 'sequence_idx': None, 'track_spatial_stride': 1, '_wandb': {}}
9
+ 2026-06-02 06:50:12,060 INFO MainThread:152969 [wandb_init.py:init():896] starting backend
10
+ 2026-06-02 06:50:12,777 INFO MainThread:152969 [wandb_init.py:init():911] sending inform_init request
11
+ 2026-06-02 06:50:13,568 INFO MainThread:152969 [wandb_init.py:init():919] backend started and connected
12
+ 2026-06-02 06:50:13,571 INFO MainThread:152969 [wandb_init.py:init():989] updated telemetry
13
+ 2026-06-02 06:50:13,661 INFO MainThread:152969 [wandb_init.py:init():1013] communicating run to backend with 90.0 second timeout
14
+ 2026-06-02 06:50:14,653 INFO MainThread:152969 [wandb_init.py:init():1058] starting run threads in backend
15
+ 2026-06-02 06:50:15,600 INFO MainThread:152969 [wandb_run.py:_console_start():2542] atexit reg
16
+ 2026-06-02 06:50:15,600 INFO MainThread:152969 [wandb_run.py:_redirect():2391] redirect: wrap_raw
17
+ 2026-06-02 06:50:15,600 INFO MainThread:152969 [wandb_run.py:_redirect():2460] Wrapping output streams.
18
+ 2026-06-02 06:50:15,600 INFO MainThread:152969 [wandb_run.py:_redirect():2483] Redirects installed.
19
+ 2026-06-02 06:50:15,631 INFO MainThread:152969 [wandb_init.py:init():1098] run started, returning control to user process
20
+ 2026-06-02 07:00:12,998 INFO MainThread:152969 [wandb_run.py:_finish():2309] finishing run nirvanalan/track4d-360/ixxjov67
21
+ 2026-06-02 07:00:13,001 INFO MainThread:152969 [wandb_run.py:_atexit_cleanup():2508] got exitcode: 0
22
+ 2026-06-02 07:00:13,002 INFO MainThread:152969 [wandb_run.py:_restore():2490] restore
23
+ 2026-06-02 07:00:13,002 INFO MainThread:152969 [wandb_run.py:_restore():2496] restore done
24
+ 2026-06-02 07:00:14,187 INFO MainThread:152969 [wandb_run.py:_footer_sync_info():3886] logging synced files
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_conv3dres0_cond_dropout/wandb/run-20260602_065012-ixxjov67/run-ixxjov67.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:555b9456a92d6560d1e50f8fe5a5f5f82fc801c9f8646cb469ca07ad7b0d41c8
3
+ size 694119
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/ismb_sbatch_14b_8node_288x512_fsdp_raymaptokens_nextstage_recammaster_pair1x49_eb32_lr2e5.sh ADDED
@@ -0,0 +1,170 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+ #SBATCH --job-name=14b_8node_288x512_pair1x49_next_eb32_lr2e5
3
+ #SBATCH --nodes=8
4
+ #SBATCH --gpus=32
5
+ #SBATCH --gpus-per-node=4
6
+ #SBATCH --ntasks-per-node=1
7
+ #SBATCH --time=24:00:00
8
+ #SBATCH --output=out/exp_%j_%x.out
9
+
10
+ # Track4D-360 14B FSDP 288x512 next-stage finetune.
11
+ #
12
+ # Main changes from the 288p backendmix OOM-debug launcher:
13
+ # - new DATA_TAG / output directory;
14
+ # - warm-start from the latest existing 288p backendmix checkpoint;
15
+ # - remove most static datasets from DATA_CONFIG;
16
+ # - add ReCamMaster CoWTracker + square512 source49 + compact pair1x49
17
+ # VGGT-Omega labels, q1-only by dataset construction.
18
+ #
19
+ # Submit from repo root on Isambard:
20
+ # sbatch bash_scripts/track4d_360/ismb/14b/sbatch/ismb_sbatch_14b_8node_288x512_fsdp_raymaptokens_nextstage_recammaster_pair1x49_eb32_lr2e5.sh
21
+
22
+ export NNODES=8
23
+ export MODEL_SIZE=14b
24
+ export CONDA_ENV="${CONDA_ENV:-th29py312}"
25
+ export HEIGHT=288
26
+ export WIDTH=512
27
+ export NUM_FRAMES=49
28
+ export main_process_port="${main_process_port:-29595}"
29
+
30
+ export WAN_T2V14B_DIR="${WAN_T2V14B_DIR:-${SLURM_SUBMIT_DIR}/weights/Wan2.1-T2V-14B}"
31
+ export VXF_INIT="${VXF_INIT:-${WAN_T2V14B_DIR}/diffusion_pytorch_model*.safetensors}"
32
+ export MODEL_ID_PREFIX="${MODEL_ID_PREFIX:-Wan-AI/Wan2.1-T2V-14B}"
33
+ export TOKENIZER_PATH="${TOKENIZER_PATH:-${WAN_T2V14B_DIR}/google/umt5-xxl}"
34
+ export T5_PATH="${T5_PATH:-${WAN_T2V14B_DIR}/models_t5_umt5-xxl-enc-bf16.pth}"
35
+ export VAE_PATH="${VAE_PATH:-${WAN_T2V14B_DIR}/Wan2.1_VAE.pth}"
36
+
37
+ # Prefer the current 288p backendmix run as the next-stage warm start. Fall
38
+ # back to the 384p backendmix/raymap-token parents only when no 288p checkpoint exists.
39
+ export BACKENDMIX_288_PARENT_DIR="${BACKENDMIX_288_PARENT_DIR:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_warmstart384_backendmix_q1to4_s1248_jitter_norecam_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_warmstart384_backendmix_q1to4_s1248_jitter_norecam_eb32_288x512_cond_dropout}"
40
+ export BACKENDMIX_384_PARENT_DIR="${BACKENDMIX_384_PARENT_DIR:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_resume_latest_backendmix_q1to4_s1248_jitter_norecam_eb32_384x672_cond_dropout_384x672x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_resume_latest_backendmix_q1to4_s1248_jitter_norecam_eb32_384x672_cond_dropout}"
41
+ export RAYMAP_384_PARENT_DIR="${RAYMAP_384_PARENT_DIR:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_real_synth_syn4d_recam_syncam_softsplat_raymaptokens_t2v_init_scratch_eb32_cond_dropout_384x672x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_real_synth_syn4d_recam_syncam_softsplat_raymaptokens_t2v_init_scratch_eb32_cond_dropout}"
42
+ # Warm-start from the 288p backendmix run's step-9000 checkpoint — the requested
43
+ # next-stage starting point. (FSDP1/FSDP2 read it identically: warm-start loads
44
+ # the safetensors BEFORE accelerator.prepare, on the unwrapped model, so it is
45
+ # distributed-backend-independent.) Override WARM_START=... to pick another;
46
+ # falls back to latest backendmix / 384p parents if step-9000 is absent.
47
+ if [ -z "${WARM_START:-}" ]; then
48
+ if [ -f "${BACKENDMIX_288_PARENT_DIR}/step-9000.safetensors" ]; then
49
+ export WARM_START="${BACKENDMIX_288_PARENT_DIR}/step-9000.safetensors"
50
+ else
51
+ latest="$(find "${BACKENDMIX_288_PARENT_DIR}" -maxdepth 1 -name 'step-*.safetensors' 2>/dev/null | sort -V | tail -1)"
52
+ if [ -z "${latest}" ]; then
53
+ latest="$(find "${BACKENDMIX_384_PARENT_DIR}" -maxdepth 1 -name 'step-*.safetensors' 2>/dev/null | sort -V | tail -1)"
54
+ fi
55
+ if [ -z "${latest}" ]; then
56
+ latest="$(find "${RAYMAP_384_PARENT_DIR}" -maxdepth 1 -name 'step-*.safetensors' 2>/dev/null | sort -V | tail -1)"
57
+ fi
58
+ export WARM_START="${latest:-${RAYMAP_384_PARENT_DIR}/step-3500.safetensors}"
59
+ fi
60
+ fi
61
+
62
+ if ! compgen -G "${VXF_INIT}" >/dev/null; then
63
+ echo "ERROR: VXF_INIT not found: ${VXF_INIT}" >&2
64
+ exit 1
65
+ fi
66
+ [ -d "${TOKENIZER_PATH}" ] || { echo "ERROR: TOKENIZER_PATH not found: ${TOKENIZER_PATH}" >&2; exit 1; }
67
+ [ -f "${T5_PATH}" ] || { echo "ERROR: T5_PATH not found: ${T5_PATH}" >&2; exit 1; }
68
+ [ -f "${VAE_PATH}" ] || { echo "ERROR: VAE_PATH not found: ${VAE_PATH}" >&2; exit 1; }
69
+
70
+ export TARGET_EFFECTIVE_BATCH=32
71
+ export gradient_accumulation_steps=1
72
+ export learning_rate="${learning_rate:-2e-5}"
73
+ export save_steps="${save_steps:-500}"
74
+ export dataset_num_workers="${dataset_num_workers:-7}"
75
+ export TRACK4D360_STOP_AFTER_NEXT_SAVE="${TRACK4D360_STOP_AFTER_NEXT_SAVE:-0}"
76
+
77
+ export dit_finetune_modules="${dit_finetune_modules:-self_attn}"
78
+ export DIT_FINETUNE_TAG="${DIT_FINETUNE_TAG:-attn}"
79
+ export use_plucker_raymap_tokens=1
80
+ export train_plucker_camera_control=0
81
+ export PLUCKER_WANDB_TAG="${PLUCKER_WANDB_TAG:-plucker_raymap_tokens}"
82
+
83
+ export DATA_CONFIG="${DATA_CONFIG:-${SLURM_SUBMIT_DIR}/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml}"
84
+ export DATA_TAG="${DATA_TAG:-mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512}"
85
+
86
+ export RECAMMASTER_SOURCE49_ROOT="${RECAMMASTER_SOURCE49_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_square512_source49}"
87
+ export RECAMMASTER_PAIR1X49_LABEL_ROOT="${RECAMMASTER_PAIR1X49_LABEL_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_pair1x49_square512_scale_labels}"
88
+ export RECAMMASTER_COW_ROOT="${RECAMMASTER_COW_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_VGGT_Omega_CoWTracker_SourceTracks_all}"
89
+ export TRACK4D360_RECAM_TRACK_SAMPLE_MODE="${TRACK4D360_RECAM_TRACK_SAMPLE_MODE:-grid}"
90
+ export TRACK4D360_RECAM_TRACK_STRIDE="${TRACK4D360_RECAM_TRACK_STRIDE:-2}"
91
+ export TRACK4D360_RECAM_TRACK_SAMPLES="${TRACK4D360_RECAM_TRACK_SAMPLES:-0}"
92
+ # Pair1x49 labels currently cover only the priority block: first 750 numeric
93
+ # train scenes inside each K/focal group (e.g. train/f18_aperture10/scene000..).
94
+ export TRACK4D360_RECAM_SCENE_OFFSET_PER_K="${TRACK4D360_RECAM_SCENE_OFFSET_PER_K:-0}"
95
+ export TRACK4D360_RECAM_SCENE_LIMIT_PER_K="${TRACK4D360_RECAM_SCENE_LIMIT_PER_K:-750}"
96
+
97
+ # Fixed post-compaction track identity count — MUST be set (the 24h/10k-step
98
+ # backendmix run had this at 65536; this launcher previously omitted it). The
99
+ # ReCamMaster CoW tracks are dense (grid stride 2, samples=0 = uncapped), so
100
+ # without a cap N varies per sample and a heavy draw spikes the track adapter's
101
+ # proj_down activation (B,N,T,hidden) -> the observed CUDA OOM at
102
+ # conv3d_track_adapter.py proj_down (1.16 GiB alloc, step ~770). Fixing N also
103
+ # pads small draws to a constant shape, eliminating per-step fragmentation. Lower
104
+ # to 32768 for extra headroom on contended nodes.
105
+ export TRACK4D360_TRACK_IDENTITY_FIXED_N="${TRACK4D360_TRACK_IDENTITY_FIXED_N:-65536}"
106
+
107
+ export TRACK4D360_SPARSE_TRACK_PROB="${TRACK4D360_SPARSE_TRACK_PROB:-1.0}"
108
+ export TRACK4D360_SPARSE_TRACK_STRIDE="${TRACK4D360_SPARSE_TRACK_STRIDE:-2|4|8}"
109
+ export TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB="${TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB:-1.0}"
110
+ export TRACK4D360_SPARSE_TRACK_BLOCK_SIZE="${TRACK4D360_SPARSE_TRACK_BLOCK_SIZE:-64}"
111
+ export TRACK4D360_MULTI_QUERY_PROB="${TRACK4D360_MULTI_QUERY_PROB:-1.0}"
112
+ export TRACK4D360_MULTI_QUERY_CHOICES="${TRACK4D360_MULTI_QUERY_CHOICES:-1|2|3|4}"
113
+ export TRACK4D360_GEOMETRY_NOISE_PROB="${TRACK4D360_GEOMETRY_NOISE_PROB:-1.0}"
114
+ export TRACK4D360_TRACK_OFFSET_STD="${TRACK4D360_TRACK_OFFSET_STD:-0.015}"
115
+ export TRACK4D360_TRACK_TEMPORAL_JITTER_STD="${TRACK4D360_TRACK_TEMPORAL_JITTER_STD:-0.006}"
116
+ export TRACK4D360_TRACK_XY_NOISE_RATIO="${TRACK4D360_TRACK_XY_NOISE_RATIO:-0.25}"
117
+ export TRACK4D360_TRACK_FRAME_DEPTH_JITTER_RATIO="${TRACK4D360_TRACK_FRAME_DEPTH_JITTER_RATIO:-0.5}"
118
+ export TRACK4D360_DEPTH_OFFSET_STD="${TRACK4D360_DEPTH_OFFSET_STD:-0.006}"
119
+ export TRACK4D360_DEPTH_SCALE_JITTER="${TRACK4D360_DEPTH_SCALE_JITTER:-0.01}"
120
+ export TRACK4D360_SOURCE_RGB_NOISE_STD="${TRACK4D360_SOURCE_RGB_NOISE_STD:-0.01}"
121
+ export TRACK4D360_SOURCE_RGB_FRAME_JITTER_STD="${TRACK4D360_SOURCE_RGB_FRAME_JITTER_STD:-0.02}"
122
+ export TRACK4D360_DYNAMIC_MASK_AUG_PROB="${TRACK4D360_DYNAMIC_MASK_AUG_PROB:-0.5}"
123
+ export TRACK4D360_DYNAMIC_MASK_MORPH_KERNEL="${TRACK4D360_DYNAMIC_MASK_MORPH_KERNEL:-5}"
124
+ export TRACK4D360_DYNAMIC_MASK_FLIP_PROB="${TRACK4D360_DYNAMIC_MASK_FLIP_PROB:-0.002}"
125
+
126
+ export dense_warp_mode="${dense_warp_mode:-hybrid_softsplat}"
127
+ export dense_warp_motion_threshold="${dense_warp_motion_threshold:-0.005}"
128
+ export dense_conditioning_noise_max_level="${dense_conditioning_noise_max_level:-0.15}"
129
+ export source_mask_boundary_depth_dropout_prob="${source_mask_boundary_depth_dropout_prob:-0.5}"
130
+ export source_mask_boundary_kernel_size="${source_mask_boundary_kernel_size:-5}"
131
+ export freeze_track4d_condition_modules="${freeze_track4d_condition_modules:-0}"
132
+
133
+ export use_gradient_checkpointing=1
134
+ export use_gradient_checkpointing_offload="${use_gradient_checkpointing_offload:-0}"
135
+ export VXF_OFFLOAD_T5_TO_CPU="${VXF_OFFLOAD_T5_TO_CPU:-1}"
136
+ export TRACK4D360_MEM_PROBE="${TRACK4D360_MEM_PROBE:-0}"
137
+ export TRACK4D360_ADAPTER_MEM_PROFILE="${TRACK4D360_ADAPTER_MEM_PROFILE:-0}"
138
+ export PYTORCH_CUDA_ALLOC_CONF="${PYTORCH_CUDA_ALLOC_CONF:-${PYTORCH_ALLOC_CONF:-expandable_segments:True}}"
139
+ export PYTORCH_ALLOC_CONF="${PYTORCH_ALLOC_CONF:-${PYTORCH_CUDA_ALLOC_CONF}}"
140
+ export TRACK4D360_DENSE_HYBRID_CACHE_FRAMES="${TRACK4D360_DENSE_HYBRID_CACHE_FRAMES:-7}"
141
+ export TRACK4D360_DENSE_HYBRID_TGT_CHUNK="${TRACK4D360_DENSE_HYBRID_TGT_CHUNK:-2}"
142
+
143
+ export NCCL_P2P_DISABLE="${NCCL_P2P_DISABLE:-0}"
144
+ export NCCL_P2P_LEVEL="${NCCL_P2P_LEVEL:-NVL}"
145
+ export NCCL_NVLS_ENABLE=0
146
+ export NCCL_NSOCKS_PERTHREAD=8
147
+ export NCCL_SOCKET_NTHREADS=8
148
+ export NCCL_MAX_NCHANNELS="${NCCL_MAX_NCHANNELS:-32}"
149
+ export NCCL_BUFFSIZE="${NCCL_BUFFSIZE:-8388608}"
150
+ export NCCL_DEBUG="${NCCL_DEBUG:-WARN}"
151
+
152
+ export TRACK4D360_CAPTION_OUTPUT_ROOT="${TRACK4D360_CAPTION_OUTPUT_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output}"
153
+ export SYN4D_CAPTION_DIR="${SYN4D_CAPTION_DIR:-${TRACK4D360_CAPTION_OUTPUT_ROOT}/Syn4D}"
154
+
155
+ echo "[raymap-tokens 14B pair1x49 next] T2V init: ${VXF_INIT}"
156
+ echo "[raymap-tokens 14B pair1x49 next] warm start: ${WARM_START}"
157
+ echo "[raymap-tokens 14B pair1x49 next] data: ${DATA_CONFIG}"
158
+ echo "[raymap-tokens 14B pair1x49 next] ReCam source49: ${RECAMMASTER_SOURCE49_ROOT}"
159
+ echo "[raymap-tokens 14B pair1x49 next] ReCam pair1x49 labels: ${RECAMMASTER_PAIR1X49_LABEL_ROOT}"
160
+ echo "[raymap-tokens 14B pair1x49 next] ReCam CoW: ${RECAMMASTER_COW_ROOT}"
161
+ echo "[raymap-tokens 14B pair1x49 next] ReCam q1 track sampling: mode=${TRACK4D360_RECAM_TRACK_SAMPLE_MODE}, stride=${TRACK4D360_RECAM_TRACK_STRIDE}, samples=${TRACK4D360_RECAM_TRACK_SAMPLES}"
162
+ echo "[raymap-tokens 14B pair1x49 next] ReCam scene window per K: offset=${TRACK4D360_RECAM_SCENE_OFFSET_PER_K}, limit=${TRACK4D360_RECAM_SCENE_LIMIT_PER_K}"
163
+ echo "[raymap-tokens 14B pair1x49 next] q policy: ReCam q1-only; synthetic choices=${TRACK4D360_MULTI_QUERY_CHOICES}"
164
+ echo "[raymap-tokens 14B pair1x49 next] sparse synthetic tracks: stride_choices=${TRACK4D360_SPARSE_TRACK_STRIDE}, block_jitter_prob=${TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB}"
165
+ echo "[raymap-tokens 14B pair1x49 next] trainables: self_attn + raymap Ecam/camera_projector + active track/dense condition modules"
166
+ echo "[raymap-tokens 14B pair1x49 next] freeze track/dense condition modules: ${freeze_track4d_condition_modules}"
167
+ echo "[raymap-tokens 14B pair1x49 next] batch: effective=32, grad_accum=1, lr=${learning_rate}, save_steps=${save_steps}"
168
+ echo "[raymap-tokens 14B pair1x49 next] memory: grad_ckpt_offload=${use_gradient_checkpointing_offload}, hybrid_cache_frames=${TRACK4D360_DENSE_HYBRID_CACHE_FRAMES}, hybrid_tgt_chunk=${TRACK4D360_DENSE_HYBRID_TGT_CHUNK}, cuda_alloc_conf=${PYTORCH_CUDA_ALLOC_CONF}"
169
+
170
+ source "${SLURM_SUBMIT_DIR}/bash_scripts/track4d_360/ismb/sbatch_common.sh"
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Next-stage 288x512 backendmix finetune with cleaned ReCamMaster pair1x49 labels.
2
+ #
3
+ # Mix: 35% ReCamMaster pair1x49, 35% Syn4D, 5% Kubric49, 5% Kubric81,
4
+ # and 20% legacy static datasets. The static 20% keeps the previous backendmix
5
+ # relative weights: DL3DV:RE10K:MVS-Synth:TartanAir = 65000:7500:12500:25000.
6
+ # ReCamMaster uses compact square512 source49 + pair1x49 VGGT-Omega scale
7
+ # labels plus CoWTracker source tracks and is q1-only by construction.
8
+ #
9
+ # sbatch_common.sh still sanity-checks for the old MVS-Synth guard string:
10
+ # exclude_sequence_ids=['0000']. This config has no MVS-Synth entry.
11
+
12
+ num_views: 49
13
+ resolution: [512, 288]
14
+
15
+ syn4d_root: ${oc.env:SYN4D_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/Syn4D}
16
+ syn4d_caption_dir: ${oc.env:SYN4D_CAPTION_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/Syn4D}
17
+ syn4d_valid_sequences: configs/syn4d_valid_sequences.txt
18
+ dl3dv_root: ${oc.env:DL3DV_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/dl3dv_tar}
19
+ re10k_root: ${oc.env:RE10K_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/processed_re10k}
20
+ mvs_synth_root: ${oc.env:MVS_SYNTH_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/processed_mvs_synth}
21
+ tartanair_root: ${oc.env:TARTANAIR_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/processed_tartanair}
22
+ kubric49_root: ${oc.env:KUBRIC49_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/track_dataset/train_preprocessed_49frames}
23
+ kubric81_root: ${oc.env:KUBRIC81_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/track_dataset/train_preprocessed_81frames}
24
+ kubric49_caption_cache_dir: ${oc.env:KUBRIC49_CAPTION_CACHE_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/kubric_train_49}
25
+ kubric81_caption_cache_dir: ${oc.env:KUBRIC81_CAPTION_CACHE_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/kubric_train_81}
26
+ recammaster_caption_cache_dir: ${oc.env:RECAMMASTER_CAPTION_CACHE_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/recammaster}
27
+ recammaster_root: ${oc.env:RECAMMASTER_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset}
28
+ multicam_fg_mask_root: ${oc.env:MULTICAM_FG_MASK_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_SAM3_FG_81f_sources_person_human_288x512}
29
+ recammaster_source49_root: ${oc.env:RECAMMASTER_SOURCE49_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_square512_source49}
30
+ recammaster_pair1x49_label_root: ${oc.env:RECAMMASTER_PAIR1X49_LABEL_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_pair1x49_square512_scale_labels}
31
+ recammaster_cow_root: ${oc.env:RECAMMASTER_COW_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_VGGT_Omega_CoWTracker_SourceTracks_all}
32
+ recammaster_track_sample_mode: ${oc.env:TRACK4D360_RECAM_TRACK_SAMPLE_MODE,grid}
33
+ recammaster_track_stride: ${oc.env:TRACK4D360_RECAM_TRACK_STRIDE,2}
34
+ recammaster_track_samples: ${oc.env:TRACK4D360_RECAM_TRACK_SAMPLES,0}
35
+ recammaster_scene_offset_per_k: ${oc.env:TRACK4D360_RECAM_SCENE_OFFSET_PER_K,0}
36
+ recammaster_scene_limit_per_k: ${oc.env:TRACK4D360_RECAM_SCENE_LIMIT_PER_K,750}
37
+
38
+ sparse_track_prob: ${oc.env:TRACK4D360_SPARSE_TRACK_PROB,1.0}
39
+ sparse_track_stride: ${oc.env:TRACK4D360_SPARSE_TRACK_STRIDE,2|4|8}
40
+ sparse_track_query_stride_mode: ${oc.env:TRACK4D360_SPARSE_TRACK_QUERY_STRIDE_MODE,query_count_floor}
41
+ sparse_track_block_jitter_prob: ${oc.env:TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB,1.0}
42
+ sparse_track_block_size: ${oc.env:TRACK4D360_SPARSE_TRACK_BLOCK_SIZE,64}
43
+ multi_query_prob: ${oc.env:TRACK4D360_MULTI_QUERY_PROB,1.0}
44
+ multi_query_choices: ${oc.env:TRACK4D360_MULTI_QUERY_CHOICES,1|2|3|4}
45
+ geometry_noise_prob: ${oc.env:TRACK4D360_GEOMETRY_NOISE_PROB,1.0}
46
+ track_offset_std: ${oc.env:TRACK4D360_TRACK_OFFSET_STD,0.015}
47
+ track_temporal_jitter_std: ${oc.env:TRACK4D360_TRACK_TEMPORAL_JITTER_STD,0.006}
48
+ track_xy_noise_ratio: ${oc.env:TRACK4D360_TRACK_XY_NOISE_RATIO,0.25}
49
+ track_frame_depth_jitter_ratio: ${oc.env:TRACK4D360_TRACK_FRAME_DEPTH_JITTER_RATIO,0.5}
50
+ depth_offset_std: ${oc.env:TRACK4D360_DEPTH_OFFSET_STD,0.006}
51
+ depth_scale_jitter: ${oc.env:TRACK4D360_DEPTH_SCALE_JITTER,0.01}
52
+ source_rgb_noise_std: ${oc.env:TRACK4D360_SOURCE_RGB_NOISE_STD,0.01}
53
+ source_rgb_frame_jitter_std: ${oc.env:TRACK4D360_SOURCE_RGB_FRAME_JITTER_STD,0.02}
54
+ dynamic_mask_aug_prob: ${oc.env:TRACK4D360_DYNAMIC_MASK_AUG_PROB,0.5}
55
+ dynamic_mask_morph_kernel: ${oc.env:TRACK4D360_DYNAMIC_MASK_MORPH_KERNEL,5}
56
+ dynamic_mask_flip_prob: ${oc.env:TRACK4D360_DYNAMIC_MASK_FLIP_PROB,0.002}
57
+
58
+ train_datasets:
59
+ # 35% ReCamMaster CoWTracker + square512 source49 + compact pair1x49 VGGT labels.
60
+ # This dataset is q1 only: it always emits track_query_indices=[0]. The
61
+ # default track density is grid stride 2 to avoid lifting every CoW pixel
62
+ # during training; override TRACK4D360_RECAM_TRACK_SAMPLE_MODE=all for dense.
63
+ - "87500 @ ReCamMasterCowVGGT_NVS(ROOT='${recammaster_root}', vggt_root='${recammaster_source49_root}', source_vggt_root='${recammaster_source49_root}', pair1x49_label_root='${recammaster_pair1x49_label_root}', geometry_source='pair1x49_label_scale_only', cow_root='${recammaster_cow_root}', fg_mask_root='${multicam_fg_mask_root}', caption_cache_dir='${recammaster_caption_cache_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=87500, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, vggt_post_crop_mode='cow_aspect', rgb_post_crop_mode='cow_aspect', track_sample_mode='${recammaster_track_sample_mode}', track_stride=${recammaster_track_stride}, track_samples=${recammaster_track_samples}, scene_offset_per_k=${recammaster_scene_offset_per_k}, scene_limit_per_k=${recammaster_scene_limit_per_k})"
64
+ # 35% Syn4D dynamic multi-camera mesh-tracked 3D.
65
+ - "87500 @ GeometryRobustnessWrapper_NVS(Syn4D_Multi_NVS(ROOT='${syn4d_root}', caption_dir='${syn4d_caption_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=87500, track_spatial_stride=1, exclude_scenes=['flying_group'], valid_sequences_file='${syn4d_valid_sequences}', use_mp4_rgb=True, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, return_dynamic_masks=True, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}'), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
66
+ # 5% Kubric 49-frame dynamic data.
67
+ - "12500 @ GeometryRobustnessWrapper_NVS(Kubric_NVS(ROOT='${kubric49_root}', caption_cache_dir='${kubric49_caption_cache_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=12500, track_spatial_stride=1, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, return_dynamic_masks=True, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}'), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
68
+ # 5% Kubric 81-frame dynamic data sampled to 49.
69
+ - "12500 @ GeometryRobustnessWrapper_NVS(Kubric_NVS(ROOT='${kubric81_root}', caption_cache_dir='${kubric81_caption_cache_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=12500, track_spatial_stride=1, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, return_dynamic_masks=True, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}'), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
70
+ # 11.82% DL3DV (real, DA3 depth). Static total is 20%; this keeps the old static relative weight.
71
+ - "29550 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${dl3dv_root}', dataset_type='dl3dv', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=29550, track_spatial_stride=1, min_overlap_ratio=0.15, min_shift_ratio=0.15, max_shift_ratio=0.5, shift_beta_ab=[2.0, 5.0], src_center_in_target_prob=0.5, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, use_da3_depth=True), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
72
+ # 1.38% RE10K (real, DA3 depth).
73
+ - "3450 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${re10k_root}', dataset_type='re10k', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=3450, track_spatial_stride=1, min_overlap_ratio=0.15, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
74
+ # 2.28% MVS-Synth (synthetic, GT depth; scene 0000 held out).
75
+ - "5700 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${mvs_synth_root}', dataset_type='mvs_synth', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=5700, track_spatial_stride=1, min_overlap_ratio=0.15, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, exclude_sequence_ids=['0000']), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
76
+ # 4.52% TartanAir (static forward-moving scenes).
77
+ - "11300 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${tartanair_root}', dataset_type='tartanair', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=11300, track_spatial_stride=1, min_overlap_ratio=0.15, min_shift_ratio=0.15, max_shift_ratio=0.5, shift_beta_ab=[2.0, 5.0], src_center_in_target_prob=1.0, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/torch_profiler/rank0.1780383278694422947.pt.trace.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5c922ef6e320bae2b92df1353ebc22a2c12ff04da2f95db04157dd3d501f2a20
3
+ size 4772489474
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/debug-internal.log ADDED
@@ -0,0 +1,87 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-06-02T06:50:49.974405605Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-06-02T06:50:49.976821599Z","level":"INFO","msg":"stream: starting","core version":"0.26.0"}
3
+ {"time":"2026-06-02T06:50:50.36107604Z","level":"INFO","msg":"stream: created new stream","id":"nn58xrpu"}
4
+ {"time":"2026-06-02T06:50:50.36114999Z","level":"INFO","msg":"handler: started"}
5
+ {"time":"2026-06-02T06:50:50.362005773Z","level":"INFO","msg":"stream: started"}
6
+ {"time":"2026-06-02T06:50:50.362019596Z","level":"INFO","msg":"writer: started","stream_id":"nn58xrpu"}
7
+ {"time":"2026-06-02T06:50:50.362095786Z","level":"INFO","msg":"sender: started"}
8
+ {"time":"2026-06-02T06:50:51.690481716Z","level":"INFO","msg":"filestream: sending request","total_files":1,"console_offset":0,"console_lines":1}
9
+ {"time":"2026-06-02T06:50:51.888011116Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
10
+ {"time":"2026-06-02T06:51:01.410733865Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":44}
11
+ {"time":"2026-06-02T06:51:01.428514695Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":168}
12
+ {"time":"2026-06-02T06:51:06.691419747Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":0,"events_lines":2,"console_offset":1,"console_lines":122,"uploaded_len":2}
13
+ {"time":"2026-06-02T06:51:06.9238499Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
14
+ {"time":"2026-06-02T06:51:21.69127157Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":2,"events_lines":2}
15
+ {"time":"2026-06-02T06:51:21.913715771Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
16
+ {"time":"2026-06-02T06:51:36.690922534Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":4,"events_lines":2}
17
+ {"time":"2026-06-02T06:51:36.965752742Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
18
+ {"time":"2026-06-02T06:51:51.692183121Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":6,"events_lines":2,"console_offset":123,"console_lines":1}
19
+ {"time":"2026-06-02T06:51:51.980180422Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
20
+ {"time":"2026-06-02T06:52:06.691301178Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":8,"events_lines":2,"console_offset":124,"console_lines":2}
21
+ {"time":"2026-06-02T06:52:07.06257193Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
22
+ {"time":"2026-06-02T06:52:21.69127678Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":0,"history_lines":2,"events_offset":10,"events_lines":2,"console_offset":122,"console_lines":1}
23
+ {"time":"2026-06-02T06:52:22.085867072Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
24
+ {"time":"2026-06-02T06:52:36.69122883Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":2,"history_lines":1,"events_offset":12,"events_lines":2,"console_offset":122,"console_lines":1}
25
+ {"time":"2026-06-02T06:52:36.896820176Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
26
+ {"time":"2026-06-02T06:52:51.691346553Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":3,"history_lines":1,"events_offset":14,"events_lines":2,"console_offset":122,"console_lines":1}
27
+ {"time":"2026-06-02T06:52:51.892989007Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
28
+ {"time":"2026-06-02T06:53:06.691797368Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":4,"history_lines":1,"events_offset":16,"events_lines":2,"console_offset":122,"console_lines":1}
29
+ {"time":"2026-06-02T06:53:06.871983169Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
30
+ {"time":"2026-06-02T06:53:21.691461565Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":5,"history_lines":1,"events_offset":18,"events_lines":2,"console_offset":122,"console_lines":1}
31
+ {"time":"2026-06-02T06:53:21.883845999Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
32
+ {"time":"2026-06-02T06:53:36.693338948Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":6,"history_lines":1,"events_offset":20,"events_lines":2,"console_offset":122,"console_lines":1}
33
+ {"time":"2026-06-02T06:53:36.916559761Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
34
+ {"time":"2026-06-02T06:53:51.69176395Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":7,"history_lines":2,"events_offset":22,"events_lines":2,"console_offset":122,"console_lines":1}
35
+ {"time":"2026-06-02T06:53:51.993609202Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
36
+ {"time":"2026-06-02T06:54:06.690851286Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":24,"events_lines":2,"console_offset":126,"console_lines":1}
37
+ {"time":"2026-06-02T06:54:07.004048439Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
38
+ {"time":"2026-06-02T06:54:21.691620646Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":26,"events_lines":2}
39
+ {"time":"2026-06-02T06:54:22.147645092Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
40
+ {"time":"2026-06-02T06:54:36.690891775Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":28,"events_lines":2}
41
+ {"time":"2026-06-02T06:54:37.065124365Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
42
+ {"time":"2026-06-02T06:54:51.692454398Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":30,"events_lines":2}
43
+ {"time":"2026-06-02T06:54:51.918197063Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
44
+ {"time":"2026-06-02T06:55:06.69053338Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":32,"events_lines":2}
45
+ {"time":"2026-06-02T06:55:06.905887225Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
46
+ {"time":"2026-06-02T06:55:21.691518Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":34,"events_lines":2,"console_offset":122,"console_lines":1}
47
+ {"time":"2026-06-02T06:55:21.947899715Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
48
+ {"time":"2026-06-02T06:55:36.691140706Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":9,"history_lines":1,"events_offset":36,"events_lines":2,"console_offset":122,"console_lines":1}
49
+ {"time":"2026-06-02T06:55:36.952030582Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
50
+ {"time":"2026-06-02T06:55:51.691789048Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":10,"history_lines":2,"events_offset":38,"events_lines":2,"console_offset":122,"console_lines":1}
51
+ {"time":"2026-06-02T06:55:51.898682683Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
52
+ {"time":"2026-06-02T06:56:06.691691779Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":12,"history_lines":1,"events_offset":40,"events_lines":2,"console_offset":122,"console_lines":1}
53
+ {"time":"2026-06-02T06:56:06.985943244Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
54
+ {"time":"2026-06-02T06:56:21.696443399Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":13,"history_lines":1,"events_offset":42,"events_lines":2,"console_offset":122,"console_lines":1}
55
+ {"time":"2026-06-02T06:56:21.906715321Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
56
+ {"time":"2026-06-02T06:56:36.691377746Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":14,"history_lines":2,"events_offset":44,"events_lines":2,"console_offset":122,"console_lines":1}
57
+ {"time":"2026-06-02T06:56:36.931509511Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
58
+ {"time":"2026-06-02T06:56:51.691747719Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":16,"history_lines":1,"events_offset":46,"events_lines":2,"console_offset":122,"console_lines":1}
59
+ {"time":"2026-06-02T06:56:51.960969812Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
60
+ {"time":"2026-06-02T06:57:06.691390544Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":17,"history_lines":1,"events_offset":48,"events_lines":2,"console_offset":122,"console_lines":1}
61
+ {"time":"2026-06-02T06:57:07.050274115Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
62
+ {"time":"2026-06-02T06:57:21.69204273Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":18,"history_lines":2,"events_offset":50,"events_lines":2,"console_offset":122,"console_lines":1}
63
+ {"time":"2026-06-02T06:57:22.186614525Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
64
+ {"time":"2026-06-02T06:57:36.691431427Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":20,"history_lines":1,"events_offset":52,"events_lines":2,"console_offset":122,"console_lines":1}
65
+ {"time":"2026-06-02T06:57:36.925377253Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
66
+ {"time":"2026-06-02T06:57:51.691892104Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":21,"history_lines":1,"events_offset":54,"events_lines":2,"console_offset":122,"console_lines":1}
67
+ {"time":"2026-06-02T06:57:51.935668512Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
68
+ {"time":"2026-06-02T06:58:06.691226293Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":22,"history_lines":2,"events_offset":56,"events_lines":2,"console_offset":122,"console_lines":1}
69
+ {"time":"2026-06-02T06:58:06.90327617Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
70
+ {"time":"2026-06-02T06:58:21.691228117Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":24,"history_lines":1,"events_offset":58,"events_lines":2,"console_offset":122,"console_lines":1}
71
+ {"time":"2026-06-02T06:58:21.922420057Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
72
+ {"time":"2026-06-02T06:58:36.691373952Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":25,"history_lines":1,"events_offset":60,"events_lines":2,"console_offset":122,"console_lines":1}
73
+ {"time":"2026-06-02T06:58:36.932965537Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
74
+ {"time":"2026-06-02T06:58:51.691882754Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":26,"history_lines":2,"events_offset":62,"events_lines":2,"console_offset":122,"console_lines":1}
75
+ {"time":"2026-06-02T06:58:51.913675918Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
76
+ {"time":"2026-06-02T06:59:06.691802803Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":28,"history_lines":1,"events_offset":64,"events_lines":2,"console_offset":122,"console_lines":1}
77
+ {"time":"2026-06-02T06:59:06.9232588Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
78
+ {"time":"2026-06-02T06:59:21.691448365Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":29,"history_lines":1,"events_offset":66,"events_lines":2,"console_offset":122,"console_lines":1}
79
+ {"time":"2026-06-02T06:59:21.938579666Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
80
+ {"time":"2026-06-02T06:59:30.473233165Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
81
+ {"time":"2026-06-02T06:59:30.474469258Z","level":"INFO","msg":"filestream: sending request","total_files":2,"console_offset":127,"console_lines":99,"uploaded_len":3,"complete":true,"exit_code":0}
82
+ {"time":"2026-06-02T06:59:30.683502844Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
83
+ {"time":"2026-06-02T06:59:30.684680347Z","level":"INFO","msg":"handler: operation stats","stats":{}}
84
+ {"time":"2026-06-02T06:59:30.689378199Z","level":"INFO","msg":"stream: closing"}
85
+ {"time":"2026-06-02T06:59:30.689457461Z","level":"INFO","msg":"handler: closed"}
86
+ {"time":"2026-06-02T06:59:30.690169153Z","level":"INFO","msg":"sender: closed"}
87
+ {"time":"2026-06-02T06:59:30.690201056Z","level":"INFO","msg":"stream: closed"}
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/debug.log ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-06-02 06:50:49,349 INFO MainThread:199010 [wandb_setup.py:_flush():81] Current SDK version is 0.26.0
2
+ 2026-06-02 06:50:49,350 INFO MainThread:199010 [wandb_setup.py:_flush():81] Configure stats pid to 199010
3
+ 2026-06-02 06:50:49,350 INFO MainThread:199010 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-06-02 06:50:49,350 INFO MainThread:199010 [wandb_init.py:setup_run_log_directory():721] Logging user logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/logs/debug.log
5
+ 2026-06-02 06:50:49,350 INFO MainThread:199010 [wandb_init.py:setup_run_log_directory():722] Logging internal logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/logs/debug-internal.log
6
+ 2026-06-02 06:50:49,350 INFO MainThread:199010 [wandb_init.py:init():848] calling init triggers
7
+ 2026-06-02 06:50:49,351 INFO MainThread:199010 [wandb_init.py:init():853] wandb.init called with sweep_config: {}
8
+ config: {'learning_rate': 2e-05, 'weight_decay': 0.01, 'num_epochs': 200, 'save_steps': 500, 'num_workers': 7, 'prefetch_factor': 2, 'dataset_base_path': None, 'dataset_metadata_path': None, 'dataset_repeat': 1, 'dataset_num_workers': 7, 'dataloader_prefetch_factor': 2, 'data_file_keys': 'image,video', 'pointmap_stats_path': None, 'model_paths': None, 'model_id_with_origin_paths': 'Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth', 'model_variant': 'vt2v', 'extra_inputs': None, 'track_xyz_normalization': 'minmax', 'track_xyz_quantile_low': 0.02, 'track_xyz_quantile_high': 0.98, 'predict_track_residual': False, 'residual_target_xyz_source': 'clean', 'residual_zero_center_normalization': False, 'residual_linear_normalization': False, 'residual_log_scale_factor': 3.0, 'stream3r_common_scale_stat': 'mean', 'explicit_motion_loss_weight': 0.0, 'explicit_motion_loss_snr_weight': True, 'explicit_motion_loss_snr_sigmoid': True, 'explicit_motion_loss_snr_sigmoid_a': 1.0, 'explicit_motion_loss_snr_sigmoid_b': -5.0, 'explicit_motion_loss_decode_checkpoint': False, 'explicit_motion_loss_decode_chunk_size': 0, 'normalize_track_with_xyz_cond_stats': False, 'prenormalized_track_video_signed_log': False, 'use_xyz_video_condition': False, 'xyz_video_condition_scale': 1.0, 'xyz_video_condition_aug_prob': 0.0, 'xyz_video_condition_noise_std': 0.0, 'xyz_video_condition_z_noise_std': 0.0, 'xyz_video_condition_frame_scale_std': 0.0, 'xyz_video_condition_dropout_prob': 0.0, 'xyz_video_condition_mode': 'xyz', 'xyz_video_condition_cfg_dropout_prob': 0.0, 'camera_condition_cfg_dropout_prob': 0.0, 'fp8_models': None, 'offload_models': None, 'v2v_concat_mode': 'width', 'v2v_channel_patch_embed_mode': 'add', 'trainable_models': 'dit.domain_embedding,dit.control_patch_embedding', 'find_unused_parameters': False, 'task': 'sft:v2v', 'drop_text_prob': 0.1, 'drop_video_prob': 0.0, 'drop_t2v_prob': None, 'drop_i2v_prob': None, 'input_video_noise_strength': 0.0, 'input_video_noise_random_max': 500, 'output_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout', 'remove_prefix_in_ckpt': 'pipe.dit.', 'save_every_n_epochs': 2, 'max_checkpoints_to_keep': 5, 'allow_existing_checkpoints': True, 'initial_step': 2530, 'max_train_steps': 2560, 'lora_base_model': None, 'lora_target_modules': 'q,k,v,o,ffn.0,ffn.2', 'lora_rank': 32, 'lora_alpha': None, 'lora_checkpoint': None, 'preset_lora_path': None, 'preset_lora_model': None, 'use_gradient_checkpointing': True, 'use_gradient_checkpointing_offload': False, 'gradient_accumulation_steps': 1, 'use_wandb': True, 'wandb_project': 'track4d-360', 'wandb_name': 'ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp', 'wandb_entity': None, 'wandb_tags': ['ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug'], 'wandb_notes': None, 'wandb_log_gradients': False, 'wandb_log_parameters': False, 'wandb_log_freq': 100, 'wandb_watch_model': False, 'debug_latents': False, 'debug_latents_every': 500, 'debug_latents_fps': 16, 'debug_latents_quality': 5, 'debug_latents_max_frames': None, 'debug_latents_outdir': 'debug_latents', 'debug_latents_vae_tiled': True, 'debug_latents_vae_tile_size': (30, 52), 'debug_latents_vae_tile_stride': (15, 26), 'debug_missing_grads': False, 'torch_distributed_debug': None, 'use_query_frame_impulse_condition': True, 'height': 288, 'width': 512, 'max_pixels': 1048576, 'num_frames': 49, 'tokenizer_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl', 'dit_finetune_modules': 'self_attn', 'track_injection_mode': 'per_block', 'track_injection_block_mode': 'concat_project', 'train_sigma_shift': 5.0, 'dummy': False, 'steps_per_epoch': 25000, 'initialize_model_on_cpu': False, 'resume_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors', 'gs_head_init_checkpoint': None, 'joint_track_prediction': False, 'joint_track_depth_prediction': False, 'joint_depth_prediction': False, 'target_depth_log_cap': 14.0, 'target_depth_norm_type': 'log_depth', 'target_depth_max_disparity': 5.0, 'vxf_init_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors', 'channel_concat_stage': None, 'use_residual_tracks': True, 'freeze_lora_after_load': False, 'skip_v2v_required_trainables': False, 'gs_only_training': False, 'freeze_track_block_injector': False, 'freeze_track4d_condition_modules': False, 'gs_head_style': 'flashworld', 'use_gs_output': False, 'gs_zero_features': False, 'gs_feat_dim': 1024, 'gs_loss_weight': 1.0, 'gs_mse_loss_weight': 1.0, 'gs_lpips_loss_weight': 0.5, 'gs_depth_loss_weight': 0.05, 'gs_abs_depth_loss_weight': 0.0, 'gs_opacity_loss_weight': 0.1, 'gs_opacity_prune_keep_frac': 1.0, 'gs_rgb_loss_weight': None, 'gs_sh_degree': 2, 'gs_num_points_per_pixel': 2, 'no_gs_latent_detach': False, 'use_gs_feedback_training': False, 'gs_scale_cap': None, 'gs_render_patch_size': None, 'gs_legacy_union_render': False, 'gs_lyra_depth_param': 'exp', 'gs_lyra_log_depth_shift': 0.0, 'gs_lyra_dnear': 0.1, 'gs_lyra_dfar': 10.0, 'gs_replay_cache_size': 0, 'gs_dynamic_mode': 'static', 'gs_track_supervision_mode': 'pointmap', 'gs_pos_loss_weight': 1.0, 'gs_dyn_depth_loss_weight': 0.5, 'gs_opa_tv_weight': 0.01, 'gs_xyz_tv_weight': 0.01, 'gs_predict_rotation': False, 'gs_disable_delta_opacity': False, 'gs_direct_pos_loss_weight': 0.0, 'gs_track_canonical_loss_weight': 0.0, 'gs_track_delta_loss_weight': 0.0, 'gs_track_final_pos_loss_weight': 0.0, 'gs_dynamic_max_t_video': 64, 'gs_assoc_track_2d_loss_weight': 0.0, 'gs_track_accel_loss_weight': 0.0, 'gs_track_loss_quantile': 0.98, 'gs_track_canonical_detach': False, 'gs_track_joint_render': False, 'cond_drop_track_prob': 0.1, 'cond_drop_dense_prob': 0.0, 'cond_drop_geo_prob': 0.1, 'use_track_null_token_fallback': False, 'use_dense_null_token_fallback': False, 'use_plucker_camera_control': True, 'enable_v2v_plucker_camera_control': True, 'train_plucker_camera_control': False, 'use_plucker_raymap_tokens': True, 'plucker_use_ray_origin': False, 'use_perlayer_camera_control': True, 'camera_injection_mode': 'projector', 'camera_plucker_encoder_mode': 'patchify', 'warped_condition_mode': 'off', 'warped_appearance_fusion': 'concat_proj', 'warped_geom_only_dense': True, 'multi_frame_warped_appearance': False, 'multi_frame_motion_threshold': 0.01, 'fixed_step_frac': None, 'fixed_num_inference_steps': 50, 'fixed_sigma': None, 'skip_diffusion_loss': False, 'use_dense_branch': True, 'dense_in_channels': 5, 'dense_proj_dim': 32, 'dense_num_residual_blocks': 0, 'dense_alpha_track': 1.0, 'use_depth_null_token': False, 'dense_conditioning_noise_max_level': 0.15, 'source_mask_boundary_depth_dropout_prob': 0.5, 'source_mask_boundary_kernel_size': 5, 'dense_warp_mode': 'hybrid_softsplat', 'dense_warp_motion_threshold': 0.005, 'dense_hybrid_disable_dynamic_masks': False, 'use_vista4d_latent_streams': False, 'use_gen3c_warp_latent_channels': False, 'disable_track_modules': False, 'dense_only_condition_modules': False, 'torch_profile_active_steps': 5, 'torch_profile_wait_steps': 3, 'torch_profile_warmup_steps': 1, 'torch_profile_repeat': 1, 'torch_profile_dir': None, 'torch_profile_record_shapes': False, 'torch_profile_memory': False, 'torch_profile_with_stack': False, 'track_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth.yaml', 'track4d360_data_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml', 'syn4d_root': '/scratch/shared/beegfs/zeren/Syn4D', 'scene': None, 'sequence_idx': None, 'track_spatial_stride': 1, '_wandb': {}}
9
+ 2026-06-02 06:50:49,351 INFO MainThread:199010 [wandb_init.py:init():896] starting backend
10
+ 2026-06-02 06:50:49,963 INFO MainThread:199010 [wandb_init.py:init():911] sending inform_init request
11
+ 2026-06-02 06:50:49,972 INFO MainThread:199010 [wandb_init.py:init():919] backend started and connected
12
+ 2026-06-02 06:50:49,975 INFO MainThread:199010 [wandb_init.py:init():989] updated telemetry
13
+ 2026-06-02 06:50:50,021 INFO MainThread:199010 [wandb_init.py:init():1013] communicating run to backend with 90.0 second timeout
14
+ 2026-06-02 06:50:50,887 INFO MainThread:199010 [wandb_init.py:init():1058] starting run threads in backend
15
+ 2026-06-02 06:50:51,673 INFO MainThread:199010 [wandb_run.py:_console_start():2542] atexit reg
16
+ 2026-06-02 06:50:51,674 INFO MainThread:199010 [wandb_run.py:_redirect():2391] redirect: wrap_raw
17
+ 2026-06-02 06:50:51,674 INFO MainThread:199010 [wandb_run.py:_redirect():2460] Wrapping output streams.
18
+ 2026-06-02 06:50:51,674 INFO MainThread:199010 [wandb_run.py:_redirect():2483] Redirects installed.
19
+ 2026-06-02 06:50:51,682 INFO MainThread:199010 [wandb_init.py:init():1098] run started, returning control to user process
20
+ 2026-06-02 06:59:29,656 INFO MainThread:199010 [wandb_run.py:_finish():2309] finishing run nirvanalan/track4d-360/nn58xrpu
21
+ 2026-06-02 06:59:29,663 INFO MainThread:199010 [wandb_run.py:_atexit_cleanup():2508] got exitcode: 0
22
+ 2026-06-02 06:59:29,664 INFO MainThread:199010 [wandb_run.py:_restore():2490] restore
23
+ 2026-06-02 06:59:29,664 INFO MainThread:199010 [wandb_run.py:_restore():2496] restore done
24
+ 2026-06-02 06:59:30,688 INFO MainThread:199010 [wandb_run.py:_footer_sync_info():3886] logging synced files
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/files/config.yaml ADDED
@@ -0,0 +1,700 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _wandb:
2
+ value:
3
+ cli_version: 0.26.0
4
+ e:
5
+ e80r6z0g9k261lnpba92qxhgslcr7ks2:
6
+ args:
7
+ - --model_id_with_origin_paths
8
+ - Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth
9
+ - --tokenizer_path
10
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl
11
+ - --task
12
+ - sft:v2v
13
+ - --model_variant
14
+ - vt2v
15
+ - --v2v_concat_mode
16
+ - width
17
+ - --height
18
+ - "288"
19
+ - --width
20
+ - "512"
21
+ - --num_frames
22
+ - "49"
23
+ - --track4d360_data_config
24
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml
25
+ - --steps_per_epoch
26
+ - "25000"
27
+ - --num_epochs
28
+ - "200"
29
+ - --learning_rate
30
+ - "2e-5"
31
+ - --gradient_accumulation_steps
32
+ - "1"
33
+ - --save_steps
34
+ - "500"
35
+ - --remove_prefix_in_ckpt
36
+ - pipe.dit.
37
+ - --trainable_models
38
+ - dit.domain_embedding,dit.control_patch_embedding
39
+ - --dit_finetune_modules
40
+ - self_attn
41
+ - --track_injection_mode
42
+ - per_block
43
+ - --track_injection_block_mode
44
+ - concat_project
45
+ - --track_config
46
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth.yaml
47
+ - --train_sigma_shift
48
+ - "5.0"
49
+ - --use_query_frame_impulse_condition
50
+ - --use_plucker_camera_control
51
+ - --enable_v2v_plucker_camera_control
52
+ - --use_plucker_raymap_tokens
53
+ - --use_perlayer_camera_control
54
+ - --camera_injection_mode
55
+ - projector
56
+ - --camera_plucker_encoder_mode
57
+ - patchify
58
+ - --use_dense_branch
59
+ - --dense_in_channels
60
+ - "5"
61
+ - --dense_proj_dim
62
+ - "32"
63
+ - --dense_num_residual_blocks
64
+ - "0"
65
+ - --dense_alpha_track
66
+ - "1.0"
67
+ - --dense_conditioning_noise_max_level
68
+ - "0.15"
69
+ - --source_mask_boundary_depth_dropout_prob
70
+ - "0.5"
71
+ - --source_mask_boundary_kernel_size
72
+ - "5"
73
+ - --dense_warp_mode
74
+ - hybrid_softsplat
75
+ - --dense_warp_motion_threshold
76
+ - "0.005"
77
+ - --cond_drop_geo_prob
78
+ - "0.1"
79
+ - --cond_drop_track_prob
80
+ - "0.1"
81
+ - --cond_drop_dense_prob
82
+ - "0.0"
83
+ - --initial_step
84
+ - "2530"
85
+ - --output_path
86
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout
87
+ - --dataset_num_workers
88
+ - "7"
89
+ - --dataloader_prefetch_factor
90
+ - "2"
91
+ - --drop_text_prob
92
+ - "0.1"
93
+ - --debug_latents_every
94
+ - "500"
95
+ - --allow_existing_checkpoints
96
+ - --wandb_project
97
+ - track4d-360
98
+ - --wandb_name
99
+ - ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp
100
+ - --wandb_tags
101
+ - ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug
102
+ - --use_wandb
103
+ - --vxf_init_checkpoint
104
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors
105
+ - --use_gradient_checkpointing
106
+ - --resume_checkpoint
107
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors
108
+ - --max_train_steps
109
+ - "2560"
110
+ - --torch_profile_active_steps
111
+ - "5"
112
+ - --torch_profile_wait_steps
113
+ - "3"
114
+ - --torch_profile_warmup_steps
115
+ - "1"
116
+ - --torch_profile_repeat
117
+ - "1"
118
+ - --no_torch_profile_record_shapes
119
+ - --no_torch_profile_memory
120
+ codePath: scripts/wan2.1_fun/train_track4d360.py
121
+ codePathLocal: scripts/wan2.1_fun/train_track4d360.py
122
+ cpu_count: 288
123
+ cpu_count_logical: 288
124
+ cudaVersion: "12.7"
125
+ disk:
126
+ /:
127
+ total: "255795724288"
128
+ used: "526450688"
129
+ email: yushi.lan@outlook.com
130
+ executable: /projects/public/u6ew/yslan_public/yslan_project/miniforge3/envs/th29py312/bin/python3.12
131
+ git:
132
+ commit: b6681ed2127310d14bc5b93ae5f1d20df8b9e907
133
+ remote: git@github.com:NIRVANALAN/track4d-360.git
134
+ gpu: NVIDIA GH200 120GB
135
+ gpu_count: 4
136
+ gpu_nvidia:
137
+ - architecture: Hopper
138
+ cudaCores: 16896
139
+ memoryTotal: "102625181696"
140
+ name: NVIDIA GH200 120GB
141
+ uuid: GPU-c78246be-bb85-931c-a8c3-f605c70f9873
142
+ - architecture: Hopper
143
+ cudaCores: 16896
144
+ memoryTotal: "102625181696"
145
+ name: NVIDIA GH200 120GB
146
+ uuid: GPU-d557559f-6b05-d9af-f672-98b04500c99e
147
+ - architecture: Hopper
148
+ cudaCores: 16896
149
+ memoryTotal: "102625181696"
150
+ name: NVIDIA GH200 120GB
151
+ uuid: GPU-fc063581-c253-951d-1735-1c26372c8d1e
152
+ - architecture: Hopper
153
+ cudaCores: 16896
154
+ memoryTotal: "102625181696"
155
+ name: NVIDIA GH200 120GB
156
+ uuid: GPU-dca24d57-11a6-3aa7-883b-6f639eb27927
157
+ host: nid010271
158
+ memory:
159
+ total: "919613407232"
160
+ os: Linux-6.4.0-150600.23.73_15.0.14-cray_shasta_c_64k-aarch64-with-glibc2.38
161
+ program: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/scripts/wan2.1_fun/train_track4d360.py
162
+ python: CPython 3.12.13
163
+ root: /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout
164
+ slurm:
165
+ cluster_name: gracehopper
166
+ conf: /var/spool/slurmd/conf-cache/slurm.conf
167
+ cpu_bind: quiet,mask_cpu:0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
168
+ cpu_bind_list: 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
169
+ cpu_bind_type: 'mask_cpu:'
170
+ cpu_bind_verbose: quiet
171
+ cpus_on_node: "288"
172
+ distribution: cyclic
173
+ gpus: "32"
174
+ gpus_on_node: "4"
175
+ gpus_per_node: "4"
176
+ gtids: "0"
177
+ job_account: brics.u6ew
178
+ job_cpus_per_node: 288(x8)
179
+ job_end_time: "1780389885"
180
+ job_gid: "1483804847"
181
+ job_gpus: 0,1,2,3
182
+ job_id: "4982941"
183
+ job_name: 14b_pair1x49_profile_ablate_denseres0_fsdp2
184
+ job_nodelist: nid[010271,010275,010290,010292,010296-010299]
185
+ job_num_nodes: "8"
186
+ job_partition: workq
187
+ job_qos: normal
188
+ job_start_time: "1780382685"
189
+ job_uid: "1483804847"
190
+ job_user: yushi.u6ew
191
+ jobid: "4982941"
192
+ launch_node_ipaddr: 10.242.18.96
193
+ localid: "0"
194
+ mpi_type: pmi2
195
+ nnodes: "8"
196
+ nodeid: "0"
197
+ nodelist: nid[010271,010275,010290,010292,010296-010299]
198
+ nprocs: "8"
199
+ ntasks: "8"
200
+ ntasks_per_node: "1"
201
+ oom_kill_step: "0"
202
+ prio_process: "0"
203
+ procid: "0"
204
+ srun_comm_host: 10.242.18.96
205
+ srun_comm_port: "41477"
206
+ step_gpus: 0,1,2,3
207
+ step_id: "0"
208
+ step_launcher_port: "41477"
209
+ step_nodelist: nid[010271,010275,010290,010292,010296-010299]
210
+ step_num_nodes: "8"
211
+ step_num_tasks: "8"
212
+ step_resv_ports: 26616-26617
213
+ step_tasks_per_node: 1(x8)
214
+ stepid: "0"
215
+ submit_dir: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360
216
+ submit_host: login40
217
+ task_pid: "198920"
218
+ tasks_per_node: 1(x8)
219
+ topology_addr: global.group4.nid010271
220
+ topology_addr_pattern: switch.switch.node
221
+ umask: "0022"
222
+ startedAt: "2026-06-02T06:50:49.340519Z"
223
+ writerId: e80r6z0g9k261lnpba92qxhgslcr7ks2
224
+ m: []
225
+ python_version: 3.12.13
226
+ t:
227
+ "1":
228
+ - 1
229
+ - 11
230
+ - 41
231
+ - 49
232
+ - 71
233
+ - 83
234
+ - 98
235
+ - 106
236
+ "2":
237
+ - 1
238
+ - 11
239
+ - 41
240
+ - 49
241
+ - 51
242
+ - 71
243
+ - 83
244
+ - 98
245
+ - 106
246
+ "3":
247
+ - 2
248
+ - 13
249
+ - 15
250
+ - 16
251
+ - 41
252
+ - 61
253
+ "4": 3.12.13
254
+ "5": 0.26.0
255
+ "6": 4.46.2
256
+ "12": 0.26.0
257
+ "13": linux-aarch64
258
+ allow_existing_checkpoints:
259
+ value: true
260
+ camera_condition_cfg_dropout_prob:
261
+ value: 0
262
+ camera_injection_mode:
263
+ value: projector
264
+ camera_plucker_encoder_mode:
265
+ value: patchify
266
+ channel_concat_stage:
267
+ value: null
268
+ cond_drop_dense_prob:
269
+ value: 0
270
+ cond_drop_geo_prob:
271
+ value: 0.1
272
+ cond_drop_track_prob:
273
+ value: 0.1
274
+ data_file_keys:
275
+ value: image,video
276
+ dataloader_prefetch_factor:
277
+ value: 2
278
+ dataset_base_path:
279
+ value: null
280
+ dataset_metadata_path:
281
+ value: null
282
+ dataset_num_workers:
283
+ value: 7
284
+ dataset_repeat:
285
+ value: 1
286
+ debug_latents:
287
+ value: false
288
+ debug_latents_every:
289
+ value: 500
290
+ debug_latents_fps:
291
+ value: 16
292
+ debug_latents_max_frames:
293
+ value: null
294
+ debug_latents_outdir:
295
+ value: debug_latents
296
+ debug_latents_quality:
297
+ value: 5
298
+ debug_latents_vae_tile_size:
299
+ value:
300
+ - 30
301
+ - 52
302
+ debug_latents_vae_tile_stride:
303
+ value:
304
+ - 15
305
+ - 26
306
+ debug_latents_vae_tiled:
307
+ value: true
308
+ debug_missing_grads:
309
+ value: false
310
+ dense_alpha_track:
311
+ value: 1
312
+ dense_conditioning_noise_max_level:
313
+ value: 0.15
314
+ dense_hybrid_disable_dynamic_masks:
315
+ value: false
316
+ dense_in_channels:
317
+ value: 5
318
+ dense_num_residual_blocks:
319
+ value: 0
320
+ dense_only_condition_modules:
321
+ value: false
322
+ dense_proj_dim:
323
+ value: 32
324
+ dense_warp_mode:
325
+ value: hybrid_softsplat
326
+ dense_warp_motion_threshold:
327
+ value: 0.005
328
+ disable_track_modules:
329
+ value: false
330
+ dit_finetune_modules:
331
+ value: self_attn
332
+ drop_i2v_prob:
333
+ value: null
334
+ drop_t2v_prob:
335
+ value: null
336
+ drop_text_prob:
337
+ value: 0.1
338
+ drop_video_prob:
339
+ value: 0
340
+ dummy:
341
+ value: false
342
+ enable_v2v_plucker_camera_control:
343
+ value: true
344
+ explicit_motion_loss_decode_checkpoint:
345
+ value: false
346
+ explicit_motion_loss_decode_chunk_size:
347
+ value: 0
348
+ explicit_motion_loss_snr_sigmoid:
349
+ value: true
350
+ explicit_motion_loss_snr_sigmoid_a:
351
+ value: 1
352
+ explicit_motion_loss_snr_sigmoid_b:
353
+ value: -5
354
+ explicit_motion_loss_snr_weight:
355
+ value: true
356
+ explicit_motion_loss_weight:
357
+ value: 0
358
+ extra_inputs:
359
+ value: null
360
+ find_unused_parameters:
361
+ value: false
362
+ fixed_num_inference_steps:
363
+ value: 50
364
+ fixed_sigma:
365
+ value: null
366
+ fixed_step_frac:
367
+ value: null
368
+ fp8_models:
369
+ value: null
370
+ freeze_lora_after_load:
371
+ value: false
372
+ freeze_track_block_injector:
373
+ value: false
374
+ freeze_track4d_condition_modules:
375
+ value: false
376
+ gradient_accumulation_steps:
377
+ value: 1
378
+ gs_abs_depth_loss_weight:
379
+ value: 0
380
+ gs_assoc_track_2d_loss_weight:
381
+ value: 0
382
+ gs_depth_loss_weight:
383
+ value: 0.05
384
+ gs_direct_pos_loss_weight:
385
+ value: 0
386
+ gs_disable_delta_opacity:
387
+ value: false
388
+ gs_dyn_depth_loss_weight:
389
+ value: 0.5
390
+ gs_dynamic_max_t_video:
391
+ value: 64
392
+ gs_dynamic_mode:
393
+ value: static
394
+ gs_feat_dim:
395
+ value: 1024
396
+ gs_head_init_checkpoint:
397
+ value: null
398
+ gs_head_style:
399
+ value: flashworld
400
+ gs_legacy_union_render:
401
+ value: false
402
+ gs_loss_weight:
403
+ value: 1
404
+ gs_lpips_loss_weight:
405
+ value: 0.5
406
+ gs_lyra_depth_param:
407
+ value: exp
408
+ gs_lyra_dfar:
409
+ value: 10
410
+ gs_lyra_dnear:
411
+ value: 0.1
412
+ gs_lyra_log_depth_shift:
413
+ value: 0
414
+ gs_mse_loss_weight:
415
+ value: 1
416
+ gs_num_points_per_pixel:
417
+ value: 2
418
+ gs_only_training:
419
+ value: false
420
+ gs_opa_tv_weight:
421
+ value: 0.01
422
+ gs_opacity_loss_weight:
423
+ value: 0.1
424
+ gs_opacity_prune_keep_frac:
425
+ value: 1
426
+ gs_pos_loss_weight:
427
+ value: 1
428
+ gs_predict_rotation:
429
+ value: false
430
+ gs_render_patch_size:
431
+ value: null
432
+ gs_replay_cache_size:
433
+ value: 0
434
+ gs_rgb_loss_weight:
435
+ value: null
436
+ gs_scale_cap:
437
+ value: null
438
+ gs_sh_degree:
439
+ value: 2
440
+ gs_track_accel_loss_weight:
441
+ value: 0
442
+ gs_track_canonical_detach:
443
+ value: false
444
+ gs_track_canonical_loss_weight:
445
+ value: 0
446
+ gs_track_delta_loss_weight:
447
+ value: 0
448
+ gs_track_final_pos_loss_weight:
449
+ value: 0
450
+ gs_track_joint_render:
451
+ value: false
452
+ gs_track_loss_quantile:
453
+ value: 0.98
454
+ gs_track_supervision_mode:
455
+ value: pointmap
456
+ gs_xyz_tv_weight:
457
+ value: 0.01
458
+ gs_zero_features:
459
+ value: false
460
+ height:
461
+ value: 288
462
+ initial_step:
463
+ value: 2530
464
+ initialize_model_on_cpu:
465
+ value: false
466
+ input_video_noise_random_max:
467
+ value: 500
468
+ input_video_noise_strength:
469
+ value: 0
470
+ joint_depth_prediction:
471
+ value: false
472
+ joint_track_depth_prediction:
473
+ value: false
474
+ joint_track_prediction:
475
+ value: false
476
+ learning_rate:
477
+ value: 2e-05
478
+ lora_alpha:
479
+ value: null
480
+ lora_base_model:
481
+ value: null
482
+ lora_checkpoint:
483
+ value: null
484
+ lora_rank:
485
+ value: 32
486
+ lora_target_modules:
487
+ value: q,k,v,o,ffn.0,ffn.2
488
+ max_checkpoints_to_keep:
489
+ value: 5
490
+ max_pixels:
491
+ value: 1048576
492
+ max_train_steps:
493
+ value: 2560
494
+ model_id_with_origin_paths:
495
+ value: Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth
496
+ model_paths:
497
+ value: null
498
+ model_variant:
499
+ value: vt2v
500
+ multi_frame_motion_threshold:
501
+ value: 0.01
502
+ multi_frame_warped_appearance:
503
+ value: false
504
+ no_gs_latent_detach:
505
+ value: false
506
+ normalize_track_with_xyz_cond_stats:
507
+ value: false
508
+ num_epochs:
509
+ value: 200
510
+ num_frames:
511
+ value: 49
512
+ num_workers:
513
+ value: 7
514
+ offload_models:
515
+ value: null
516
+ output_path:
517
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout
518
+ plucker_use_ray_origin:
519
+ value: false
520
+ pointmap_stats_path:
521
+ value: null
522
+ predict_track_residual:
523
+ value: false
524
+ prefetch_factor:
525
+ value: 2
526
+ prenormalized_track_video_signed_log:
527
+ value: false
528
+ preset_lora_model:
529
+ value: null
530
+ preset_lora_path:
531
+ value: null
532
+ remove_prefix_in_ckpt:
533
+ value: pipe.dit.
534
+ residual_linear_normalization:
535
+ value: false
536
+ residual_log_scale_factor:
537
+ value: 3
538
+ residual_target_xyz_source:
539
+ value: clean
540
+ residual_zero_center_normalization:
541
+ value: false
542
+ resume_checkpoint:
543
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors
544
+ save_every_n_epochs:
545
+ value: 2
546
+ save_steps:
547
+ value: 500
548
+ scene:
549
+ value: null
550
+ sequence_idx:
551
+ value: null
552
+ skip_diffusion_loss:
553
+ value: false
554
+ skip_v2v_required_trainables:
555
+ value: false
556
+ source_mask_boundary_depth_dropout_prob:
557
+ value: 0.5
558
+ source_mask_boundary_kernel_size:
559
+ value: 5
560
+ steps_per_epoch:
561
+ value: 25000
562
+ stream3r_common_scale_stat:
563
+ value: mean
564
+ syn4d_root:
565
+ value: /scratch/shared/beegfs/zeren/Syn4D
566
+ target_depth_log_cap:
567
+ value: 14
568
+ target_depth_max_disparity:
569
+ value: 5
570
+ target_depth_norm_type:
571
+ value: log_depth
572
+ task:
573
+ value: sft:v2v
574
+ tokenizer_path:
575
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl
576
+ torch_distributed_debug:
577
+ value: null
578
+ torch_profile_active_steps:
579
+ value: 5
580
+ torch_profile_dir:
581
+ value: null
582
+ torch_profile_memory:
583
+ value: false
584
+ torch_profile_record_shapes:
585
+ value: false
586
+ torch_profile_repeat:
587
+ value: 1
588
+ torch_profile_wait_steps:
589
+ value: 3
590
+ torch_profile_warmup_steps:
591
+ value: 1
592
+ torch_profile_with_stack:
593
+ value: false
594
+ track_config:
595
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth.yaml
596
+ track_injection_block_mode:
597
+ value: concat_project
598
+ track_injection_mode:
599
+ value: per_block
600
+ track_spatial_stride:
601
+ value: 1
602
+ track_xyz_normalization:
603
+ value: minmax
604
+ track_xyz_quantile_high:
605
+ value: 0.98
606
+ track_xyz_quantile_low:
607
+ value: 0.02
608
+ track4d360_data_config:
609
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml
610
+ train_plucker_camera_control:
611
+ value: false
612
+ train_sigma_shift:
613
+ value: 5
614
+ trainable_models:
615
+ value: dit.domain_embedding,dit.control_patch_embedding
616
+ use_dense_branch:
617
+ value: true
618
+ use_dense_null_token_fallback:
619
+ value: false
620
+ use_depth_null_token:
621
+ value: false
622
+ use_gen3c_warp_latent_channels:
623
+ value: false
624
+ use_gradient_checkpointing:
625
+ value: true
626
+ use_gradient_checkpointing_offload:
627
+ value: false
628
+ use_gs_feedback_training:
629
+ value: false
630
+ use_gs_output:
631
+ value: false
632
+ use_perlayer_camera_control:
633
+ value: true
634
+ use_plucker_camera_control:
635
+ value: true
636
+ use_plucker_raymap_tokens:
637
+ value: true
638
+ use_query_frame_impulse_condition:
639
+ value: true
640
+ use_residual_tracks:
641
+ value: true
642
+ use_track_null_token_fallback:
643
+ value: false
644
+ use_vista4d_latent_streams:
645
+ value: false
646
+ use_wandb:
647
+ value: true
648
+ use_xyz_video_condition:
649
+ value: false
650
+ v2v_channel_patch_embed_mode:
651
+ value: add
652
+ v2v_concat_mode:
653
+ value: width
654
+ vxf_init_checkpoint:
655
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors
656
+ wandb_entity:
657
+ value: null
658
+ wandb_log_freq:
659
+ value: 100
660
+ wandb_log_gradients:
661
+ value: false
662
+ wandb_log_parameters:
663
+ value: false
664
+ wandb_name:
665
+ value: ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp
666
+ wandb_notes:
667
+ value: null
668
+ wandb_project:
669
+ value: track4d-360
670
+ wandb_tags:
671
+ value:
672
+ - ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug
673
+ wandb_watch_model:
674
+ value: false
675
+ warped_appearance_fusion:
676
+ value: concat_proj
677
+ warped_condition_mode:
678
+ value: "off"
679
+ warped_geom_only_dense:
680
+ value: true
681
+ weight_decay:
682
+ value: 0.01
683
+ width:
684
+ value: 512
685
+ xyz_video_condition_aug_prob:
686
+ value: 0
687
+ xyz_video_condition_cfg_dropout_prob:
688
+ value: 0
689
+ xyz_video_condition_dropout_prob:
690
+ value: 0
691
+ xyz_video_condition_frame_scale_std:
692
+ value: 0
693
+ xyz_video_condition_mode:
694
+ value: xyz
695
+ xyz_video_condition_noise_std:
696
+ value: 0
697
+ xyz_video_condition_scale:
698
+ value: 1
699
+ xyz_video_condition_z_noise_std:
700
+ value: 0
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/files/output.log ADDED
@@ -0,0 +1,226 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ✓ WandB initialized: https://wandb.ai/nirvanalan/track4d-360/runs/nn58xrpu
2
+ /projects/public/u6ew/yslan_public/yslan_project/miniforge3/envs/th29py312/lib/python3.12/site-packages/accelerate/utils/fsdp_utils.py:742: UserWarning: FSDP upcast of low precision parameters to fp32 (since mixed_precision != 'no') may affect the precision of model checkpoints. This effects 589 parameters: ['pipe.dit.domain_embedding', 'pipe.dit.blocks.0.self_attn.q.weight', 'pipe.dit.blocks.0.self_attn.q.bias', 'pipe.dit.blocks.0.self_attn.k.weight', 'pipe.dit.blocks.0.self_attn.k.bias', 'pipe.dit.blocks.0.self_attn.v.weight', 'pipe.dit.blocks.0.self_attn.v.bias', 'pipe.dit.blocks.0.self_attn.o.weight', 'pipe.dit.blocks.0.self_attn.o.bias', 'pipe.dit.blocks.0.self_attn.norm_q.weight', 'pipe.dit.blocks.0.self_attn.norm_k.weight', 'pipe.dit.blocks.0.camera_projector.weight', 'pipe.dit.blocks.0.camera_projector.bias', 'pipe.dit.blocks.1.self_attn.q.weight', 'pipe.dit.blocks.1.self_attn.q.bias', 'pipe.dit.blocks.1.self_attn.k.weight', 'pipe.dit.blocks.1.self_attn.k.bias', 'pipe.dit.blocks.1.self_attn.v.weight', 'pipe.dit.blocks.1.self_attn.v.bias', 'pipe.dit.blocks.1.self_attn.o.weight', 'pipe.dit.blocks.1.self_attn.o.bias', 'pipe.dit.blocks.1.self_attn.norm_q.weight', 'pipe.dit.blocks.1.self_attn.norm_k.weight', 'pipe.dit.blocks.1.camera_projector.weight', 'pipe.dit.blocks.1.camera_projector.bias', 'pipe.dit.blocks.2.self_attn.q.weight', 'pipe.dit.blocks.2.self_attn.q.bias', 'pipe.dit.blocks.2.self_attn.k.weight', 'pipe.dit.blocks.2.self_attn.k.bias', 'pipe.dit.blocks.2.self_attn.v.weight', 'pipe.dit.blocks.2.self_attn.v.bias', 'pipe.dit.blocks.2.self_attn.o.weight', 'pipe.dit.blocks.2.self_attn.o.bias', 'pipe.dit.blocks.2.self_attn.norm_q.weight', 'pipe.dit.blocks.2.self_attn.norm_k.weight', 'pipe.dit.blocks.2.camera_projector.weight', 'pipe.dit.blocks.2.camera_projector.bias', 'pipe.dit.blocks.3.self_attn.q.weight', 'pipe.dit.blocks.3.self_attn.q.bias', 'pipe.dit.blocks.3.self_attn.k.weight', 'pipe.dit.blocks.3.self_attn.k.bias', 'pipe.dit.blocks.3.self_attn.v.weight', 'pipe.dit.blocks.3.self_attn.v.bias', 'pipe.dit.blocks.3.self_attn.o.weight', 'pipe.dit.blocks.3.self_attn.o.bias', 'pipe.dit.blocks.3.self_attn.norm_q.weight', 'pipe.dit.blocks.3.self_attn.norm_k.weight', 'pipe.dit.blocks.3.camera_projector.weight', 'pipe.dit.blocks.3.camera_projector.bias', 'pipe.dit.blocks.4.self_attn.q.weight', 'pipe.dit.blocks.4.self_attn.q.bias', 'pipe.dit.blocks.4.self_attn.k.weight', 'pipe.dit.blocks.4.self_attn.k.bias', 'pipe.dit.blocks.4.self_attn.v.weight', 'pipe.dit.blocks.4.self_attn.v.bias', 'pipe.dit.blocks.4.self_attn.o.weight', 'pipe.dit.blocks.4.self_attn.o.bias', 'pipe.dit.blocks.4.self_attn.norm_q.weight', 'pipe.dit.blocks.4.self_attn.norm_k.weight', 'pipe.dit.blocks.4.camera_projector.weight', 'pipe.dit.blocks.4.camera_projector.bias', 'pipe.dit.blocks.5.self_attn.q.weight', 'pipe.dit.blocks.5.self_attn.q.bias', 'pipe.dit.blocks.5.self_attn.k.weight', 'pipe.dit.blocks.5.self_attn.k.bias', 'pipe.dit.blocks.5.self_attn.v.weight', 'pipe.dit.blocks.5.self_attn.v.bias', 'pipe.dit.blocks.5.self_attn.o.weight', 'pipe.dit.blocks.5.self_attn.o.bias', 'pipe.dit.blocks.5.self_attn.norm_q.weight', 'pipe.dit.blocks.5.self_attn.norm_k.weight', 'pipe.dit.blocks.5.camera_projector.weight', 'pipe.dit.blocks.5.camera_projector.bias', 'pipe.dit.blocks.6.self_attn.q.weight', 'pipe.dit.blocks.6.self_attn.q.bias', 'pipe.dit.blocks.6.self_attn.k.weight', 'pipe.dit.blocks.6.self_attn.k.bias', 'pipe.dit.blocks.6.self_attn.v.weight', 'pipe.dit.blocks.6.self_attn.v.bias', 'pipe.dit.blocks.6.self_attn.o.weight', 'pipe.dit.blocks.6.self_attn.o.bias', 'pipe.dit.blocks.6.self_attn.norm_q.weight', 'pipe.dit.blocks.6.self_attn.norm_k.weight', 'pipe.dit.blocks.6.camera_projector.weight', 'pipe.dit.blocks.6.camera_projector.bias', 'pipe.dit.blocks.7.self_attn.q.weight', 'pipe.dit.blocks.7.self_attn.q.bias', 'pipe.dit.blocks.7.self_attn.k.weight', 'pipe.dit.blocks.7.self_attn.k.bias', 'pipe.dit.blocks.7.self_attn.v.weight', 'pipe.dit.blocks.7.self_attn.v.bias', 'pipe.dit.blocks.7.self_attn.o.weight', 'pipe.dit.blocks.7.self_attn.o.bias', 'pipe.dit.block
3
+ warnings.warn(
4
+
5
+ ================================================================================
6
+ TRAINABLE PARAMETERS VERIFICATION
7
+ ================================================================================
8
+
9
+ Trainable modules (94 module groups):
10
+
11
+ 📦 pipe.dit.blocks.* — 40 blocks [0..39]:
12
+ Params per block: 131,107,840 × 40 = 5,244,313,600
13
+ Submodules (example from block 0):
14
+ • self_attn: 104,888,320
15
+ • camera_projector: 26,219,520
16
+
17
+ 📦 pipe.dense_target_control_encoder.conv3d_patchify:
18
+ Total params: 10,272
19
+ ├─ weight: 10,240
20
+ ├─ bias: 32
21
+
22
+ 📦 pipe.dense_target_control_encoder.null_token:
23
+ Total params: 156
24
+ pipe.dense_target_control_encoder.null_token: 156
25
+
26
+ 📦 pipe.dense_target_control_encoder.proj_up:
27
+ Total params: 5,148
28
+ ├─ weight: 4,992
29
+ ├─ bias: 156
30
+
31
+ 📦 pipe.dit._camera_patchify_conv:
32
+ Total params: 36,870
33
+ ├─ weight: 36,864
34
+ ├─ bias: 6
35
+
36
+ 📦 pipe.dit._camera_plucker_patchify_encoder:
37
+ Total params: 35,840
38
+ ├─ weight: 30,720
39
+ ├─ bias: 5,120
40
+
41
+ 📦 pipe.dit.control_patch_embedding:
42
+ Total params: 332,800
43
+ ├─ weight: 327,680
44
+ ├─ bias: 5,120
45
+
46
+ 📦 pipe.dit.domain_embedding:
47
+ Total params: 10,240
48
+ pipe.dit.domain_embedding: 10,240
49
+
50
+ 📦 pipe.track_adapter.conv3d_patchify:
51
+ Total params: 65,568
52
+ ├─ weight: 65,536
53
+ ├─ bias: 32
54
+
55
+ 📦 pipe.track_adapter.mlp:
56
+ Total params: 80,028
57
+ ├─ 0.weight: 30,888
58
+ ├─ 2.weight: 24,336
59
+ ├─ 4.weight: 24,336
60
+ ├─ 0.bias: 156
61
+ ├─ 2.bias: 156
62
+ ├─ 4.bias: 156
63
+
64
+ 📦 pipe.track_adapter.null_token:
65
+ Total params: 156
66
+ pipe.track_adapter.null_token: 156
67
+
68
+ 📦 pipe.track_adapter.proj_down:
69
+ Total params: 5,024
70
+ ├─ weight: 4,992
71
+ ├─ bias: 32
72
+
73
+ 📦 pipe.track_adapter.proj_up:
74
+ Total params: 5,148
75
+ ├─ weight: 4,992
76
+ ├─ bias: 156
77
+
78
+ 📦 pipe.track_adapter.residual_blocks:
79
+ Total params: 55,360
80
+ ├─ 0.conv1.weight: 27,648
81
+ ├─ 0.conv2.weight: 27,648
82
+ ├─ 0.conv1.bias: 32
83
+ ├─ 0.conv2.bias: 32
84
+
85
+ 📦 pipe.track_block_injector.encoders:
86
+ Total params: 1,080,729,600
87
+ ├─ 0.weight: 27,013,120
88
+ ├─ 1.weight: 27,013,120
89
+ ├─ 10.weight: 27,013,120
90
+ ├─ 11.weight: 27,013,120
91
+ ├─ 12.weight: 27,013,120
92
+ ├─ 13.weight: 27,013,120
93
+ ├─ 14.weight: 27,013,120
94
+ ├─ 15.weight: 27,013,120
95
+ ├─ 16.weight: 27,013,120
96
+ ├─ 17.weight: 27,013,120
97
+ ├─ 18.weight: 27,013,120
98
+ ├─ 19.weight: 27,013,120
99
+ ├─ 2.weight: 27,013,120
100
+ ├─ 20.weight: 27,013,120
101
+ ├─ 21.weight: 27,013,120
102
+ ├─ 22.weight: 27,013,120
103
+ ├─ 23.weight: 27,013,120
104
+ ├─ 24.weight: 27,013,120
105
+ ├─ 25.weight: 27,013,120
106
+ ├─ 26.weight: 27,013,120
107
+ ├─ 27.weight: 27,013,120
108
+ ├─ 28.weight: 27,013,120
109
+ ├─ 29.weight: 27,013,120
110
+ ├─ 3.weight: 27,013,120
111
+ ├─ 30.weight: 27,013,120
112
+ ├─ … (55 more)
113
+
114
+ ────────────────────────────────────────────────────────────────────────────────
115
+ NOTE: FSDP is active; counts below are from the rank-local post-wrap parameter view, not a full global parameter census.
116
+ Total trainable parameters: 6,325,685,810 (6325.69M)
117
+ Total frozen parameters: 10,363,571,200 (10363.57M)
118
+ Percentage trainable (overall): 37.90%
119
+ Percentage trainable (in requested modules: dit): 33.60%
120
+ ================================================================================
121
+
122
+ [torch-profiler] rank0 enabled (wait=3, warmup=1, active=5, repeat=1) -> /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/torch_profiler
123
+ Epoch 1/200: 0%| | 29/7813 [08:13<36:46:56, 17.01s/it, opt_step=2559, loss=0.0681]
124
+ [Track4D-360] First train batch runtime flags: use_gradient_checkpointing=True, use_gradient_checkpointing_offload=False
125
+ [track4d-runtime] rank0 patched DiTBlock.forward saw mode=track block_idx=0
126
+ [Attention] Using: Flash Attention 2 (flash_attn)
127
+ [soft_splat_warp] WARN: tracks grid 160x294 (N=47040) != image grid 288x512 (N=147456); bilinear-resampling tracks_world to image grid. This warning fires once per process.
128
+
129
+ [Training] Reached max_train_steps=2560. Stopping training.
130
+
131
+ [Checkpoint] Saving 589 parameters (total 6,325,685,810 ≈ 6325.69M):
132
+
133
+ 📦 blocks.* — 40 blocks [0..39]:
134
+ Params per block: 131,107,840 × 40 = 5,244,313,600
135
+ Submodules (example from block 0):
136
+ • self_attn: 104,888,320
137
+ • camera_projector: 26,219,520
138
+
139
+ 📦 _camera_patchify_conv.bias:
140
+ Total params: 6
141
+ _camera_patchify_conv.bias: 6
142
+
143
+ 📦 _camera_patchify_conv.weight:
144
+ Total params: 36,864
145
+ _camera_patchify_conv.weight: 36,864
146
+
147
+ 📦 _camera_plucker_patchify_encoder.bias:
148
+ Total params: 5,120
149
+ _camera_plucker_patchify_encoder.bias: 5,120
150
+
151
+ 📦 _camera_plucker_patchify_encoder.weight:
152
+ Total params: 30,720
153
+ _camera_plucker_patchify_encoder.weight: 30,720
154
+
155
+ 📦 control_patch_embedding.bias:
156
+ Total params: 5,120
157
+ control_patch_embedding.bias: 5,120
158
+
159
+ 📦 control_patch_embedding.weight:
160
+ Total params: 327,680
161
+ control_patch_embedding.weight: 327,680
162
+
163
+ 📦 domain_embedding:
164
+ Total params: 10,240
165
+ domain_embedding: 10,240
166
+
167
+ 📦 pipe.dense_target_control_encoder.conv3d_patchify:
168
+ Total params: 10,272
169
+ ├─ weight: 10,240
170
+ ├─ bias: 32
171
+
172
+ 📦 pipe.dense_target_control_encoder.null_token:
173
+ Total params: 156
174
+ pipe.dense_target_control_encoder.null_token: 156
175
+
176
+ 📦 pipe.dense_target_control_encoder.proj_up:
177
+ Total params: 5,148
178
+ ├─ weight: 4,992
179
+ ├─ bias: 156
180
+
181
+ 📦 pipe.track_adapter.conv3d_patchify:
182
+ Total params: 65,568
183
+ ├─ weight: 65,536
184
+ ├─ bias: 32
185
+
186
+ 📦 pipe.track_adapter.mlp:
187
+ Total params: 80,028
188
+ ├─ 0.weight: 30,888
189
+ ├─ 2.weight: 24,336
190
+ ├─ 4.weight: 24,336
191
+ ├─ 0.bias: 156
192
+ ├─ 2.bias: 156
193
+ ├─ 4.bias: 156
194
+
195
+ 📦 pipe.track_adapter.null_token:
196
+ Total params: 156
197
+ pipe.track_adapter.null_token: 156
198
+
199
+ 📦 pipe.track_adapter.proj_down:
200
+ Total params: 5,024
201
+ ├─ weight: 4,992
202
+ ├─ bias: 32
203
+
204
+ 📦 pipe.track_adapter.proj_up:
205
+ Total params: 5,148
206
+ ├─ weight: 4,992
207
+ ├─ bias: 156
208
+
209
+ 📦 pipe.track_adapter.residual_blocks:
210
+ Total params: 55,360
211
+ ├─ 0.conv1.weight: 27,648
212
+ ├─ 0.conv2.weight: 27,648
213
+ ├─ 0.conv1.bias: 32
214
+ ├─ 0.conv2.bias: 32
215
+
216
+ 📦 pipe.track_block_injector.encoders:
217
+ Total params: 1,080,729,600
218
+ ├─ 0.weight: 27,013,120
219
+ ├─ 1.weight: 27,013,120
220
+ ├─ 10.weight: 27,013,120
221
+ ├─ 11.weight: 27,013,120
222
+ ├─ 12.weight: 27,013,120
223
+ ├─ 13.weight: 27,013,120
224
+ ├─ 14.weight: 27,013,120
225
+ ├─ 15.weight: 27,013,120
226
+ … (72 more)
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/files/requirements.txt ADDED
@@ -0,0 +1,208 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ videox-fun==1.0.1
2
+ track4d-360==0.1.0
3
+ func_timeout==4.3.5
4
+ click==8.3.2
5
+ pip==26.0.1
6
+ nvidia-cuda-nvrtc-cu12==12.6.77
7
+ ImageIO==2.37.3
8
+ numpy==1.26.4
9
+ boto3==1.42.91
10
+ albucore==0.0.24
11
+ lpips==0.1.4
12
+ torchsde==0.2.6
13
+ tomesd==0.1.3
14
+ yarl==1.23.0
15
+ typer==0.24.1
16
+ triton==3.5.0
17
+ cycler==0.12.1
18
+ multidict==6.7.1
19
+ OpenEXR==3.4.10
20
+ soxr==1.0.0
21
+ kiwisolver==1.5.0
22
+ simsimd==6.5.16
23
+ parso==0.8.6
24
+ einops==0.8.2
25
+ ftfy==6.3.1
26
+ platformdirs==4.9.6
27
+ groovy==0.1.2
28
+ nvidia-nvtx-cu12==12.6.77
29
+ sentry-sdk==2.58.0
30
+ certifi==2026.2.25
31
+ markdown-it-py==4.0.0
32
+ uvicorn==0.38.0
33
+ nvidia-cudnn-cu12==9.10.2.21
34
+ natsort==8.4.0
35
+ awscrt==0.31.2
36
+ accelerate==1.13.0
37
+ deepspeed==0.18.9
38
+ matplotlib==3.10.8
39
+ mdurl==0.1.2
40
+ regex==2026.4.4
41
+ nvidia-cuda-cupti-cu12==12.6.80
42
+ shellingham==1.5.4
43
+ joblib==1.5.3
44
+ zipp==3.23.1
45
+ ipython_pygments_lexers==1.1.1
46
+ executing==2.2.1
47
+ safetensors==0.7.0
48
+ typing-inspection==0.4.2
49
+ httpcore==1.0.9
50
+ semantic-version==2.10.0
51
+ wandb==0.26.0
52
+ asttokens==3.0.1
53
+ datasets==4.8.4
54
+ decorator==5.2.1
55
+ attrs==26.1.0
56
+ six==1.17.0
57
+ importlib_metadata==9.0.0
58
+ py-cpuinfo==9.0.0
59
+ requests==2.33.1
60
+ xxhash==3.6.0
61
+ pytorch-lightning==2.6.1
62
+ fsspec==2026.2.0
63
+ albumentations==2.0.8
64
+ soundfile==0.13.1
65
+ torchaudio==2.9.0
66
+ torchmetrics==1.9.0
67
+ nvidia-cusolver-cu12==11.7.1.2
68
+ scipy==1.17.1
69
+ jmespath==1.1.0
70
+ threadpoolctl==3.6.0
71
+ plyfile==1.1.2
72
+ peft==0.19.1
73
+ pycparser==3.0
74
+ Markdown==3.10.2
75
+ aiosignal==1.4.0
76
+ pydantic==2.13.2
77
+ decord2==3.3.0
78
+ safehttpx==0.1.7
79
+ scikit-image==0.26.0
80
+ fonttools==4.62.1
81
+ lazy-loader==0.5
82
+ smmap==5.0.3
83
+ starlette==1.0.0
84
+ Werkzeug==3.1.8
85
+ propcache==0.4.1
86
+ tensorboard==2.20.0
87
+ rich==15.0.0
88
+ modelscope==1.36.3
89
+ absl-py==2.4.0
90
+ traitlets==5.14.3
91
+ networkx==3.6.1
92
+ hf-xet==1.4.3
93
+ h5py==3.16.0
94
+ PyYAML==6.0.3
95
+ Jinja2==3.1.6
96
+ nvidia-cusparse-cu12==12.5.4.2
97
+ idna==3.11
98
+ transformers==4.46.2
99
+ httpx==0.28.1
100
+ MarkupSafe==3.0.3
101
+ tensorboard-data-server==0.7.2
102
+ pooch==1.9.0
103
+ torch==2.9.0+cu126
104
+ soupsieve==2.8.3
105
+ numba==0.65.0
106
+ torchdiffeq==0.2.5
107
+ jedi==0.19.2
108
+ charset-normalizer==3.4.7
109
+ annotated-doc==0.0.4
110
+ gradio==6.12.0
111
+ Pygments==2.20.0
112
+ pandas==3.0.2
113
+ stack-data==0.6.3
114
+ contourpy==1.3.3
115
+ tomlkit==0.14.0
116
+ scikit-learn==1.8.0
117
+ librosa==0.11.0
118
+ nvidia-cusparselt-cu12==0.7.1
119
+ pytz==2026.1.post1
120
+ go-with-the-track==0.1.0
121
+ yunchang==0.6.4
122
+ onnxruntime==1.24.4
123
+ ninja==1.13.0
124
+ sentencepiece==0.2.1
125
+ pexpect==4.9.0
126
+ multiprocess==0.70.19
127
+ omegaconf==2.3.0
128
+ h11==0.16.0
129
+ pydub==0.25.1
130
+ nvidia-cufft-cu12==11.3.0.4
131
+ xfuser==0.4.5
132
+ tifffile==2026.3.3
133
+ msgpack==1.1.2
134
+ psutil==7.2.2
135
+ ipython==9.12.0
136
+ hf-gradio==0.4.1
137
+ diffusers==0.33.1
138
+ python-dateutil==2.9.0.post0
139
+ orjson==3.11.8
140
+ huggingface_hub==0.36.2
141
+ nvidia-cuda-runtime-cu12==12.6.77
142
+ beautifulsoup4==4.14.3
143
+ ptyprocess==0.7.0
144
+ protobuf==7.34.1
145
+ GitPython==3.1.46
146
+ sympy==1.14.0
147
+ packaging==26.1
148
+ aiohttp==3.13.5
149
+ nvidia-nvshmem-cu12==3.3.20
150
+ annotated-types==0.7.0
151
+ antlr4-python3-runtime==4.9.3
152
+ gradio_client==2.4.1
153
+ stringzilla==4.6.0
154
+ flatbuffers==25.12.19
155
+ filelock==3.28.0
156
+ imageio-ffmpeg==0.6.0
157
+ fastapi==0.136.0
158
+ nvidia-nvjitlink-cu12==12.6.85
159
+ s3transfer==0.16.0
160
+ python-multipart==0.0.26
161
+ DistVAE==0.0.0b5
162
+ urllib3==2.6.3
163
+ nvidia-nccl-cu12==2.27.5
164
+ av==15.1.0
165
+ pure_eval==0.2.3
166
+ wheel==0.46.3
167
+ dill==0.4.1
168
+ flash_attn==2.8.3
169
+ track4d-360==0.1.0
170
+ opencv-python==4.11.0.86
171
+ anyio==4.13.0
172
+ pytest==9.0.3
173
+ nvidia-cufile-cu12==1.11.1.6
174
+ lightning-utilities==0.15.3
175
+ pluggy==1.6.0
176
+ grpcio==1.80.0
177
+ cffi==2.0.0
178
+ iniconfig==2.3.0
179
+ videox-fun==1.0.1
180
+ matplotlib-inline==0.2.1
181
+ tokenizers==0.20.3
182
+ pydantic_core==2.46.2
183
+ aiohappyeyeballs==2.6.1
184
+ typing_extensions==4.15.0
185
+ timm==1.0.26
186
+ pyarrow==23.0.1
187
+ mpmath==1.3.0
188
+ prompt_toolkit==3.0.52
189
+ mediapy==1.2.6
190
+ llvmlite==0.47.0
191
+ opencv-python-headless==4.11.0.86
192
+ hjson==3.1.0
193
+ tqdm==4.67.3
194
+ torchvision==0.24.0
195
+ pyparsing==3.3.2
196
+ nvidia-cublas-cu12==12.6.4.1
197
+ botocore==1.42.91
198
+ nanobind==2.9.2
199
+ wcwidth==0.6.0
200
+ gitdb==4.0.12
201
+ frozenlist==1.8.0
202
+ brotli==1.2.0
203
+ trampoline==0.1.2
204
+ pillow==12.2.0
205
+ audioread==3.1.0
206
+ lightning==2.6.1
207
+ setuptools==82.0.1
208
+ nvidia-curand-cu12==10.3.7.77
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/files/wandb-metadata.json ADDED
@@ -0,0 +1,235 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-6.4.0-150600.23.73_15.0.14-cray_shasta_c_64k-aarch64-with-glibc2.38",
3
+ "python": "CPython 3.12.13",
4
+ "startedAt": "2026-06-02T06:50:49.340519Z",
5
+ "args": [
6
+ "--model_id_with_origin_paths",
7
+ "Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth",
8
+ "--tokenizer_path",
9
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl",
10
+ "--task",
11
+ "sft:v2v",
12
+ "--model_variant",
13
+ "vt2v",
14
+ "--v2v_concat_mode",
15
+ "width",
16
+ "--height",
17
+ "288",
18
+ "--width",
19
+ "512",
20
+ "--num_frames",
21
+ "49",
22
+ "--track4d360_data_config",
23
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml",
24
+ "--steps_per_epoch",
25
+ "25000",
26
+ "--num_epochs",
27
+ "200",
28
+ "--learning_rate",
29
+ "2e-5",
30
+ "--gradient_accumulation_steps",
31
+ "1",
32
+ "--save_steps",
33
+ "500",
34
+ "--remove_prefix_in_ckpt",
35
+ "pipe.dit.",
36
+ "--trainable_models",
37
+ "dit.domain_embedding,dit.control_patch_embedding",
38
+ "--dit_finetune_modules",
39
+ "self_attn",
40
+ "--track_injection_mode",
41
+ "per_block",
42
+ "--track_injection_block_mode",
43
+ "concat_project",
44
+ "--track_config",
45
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth.yaml",
46
+ "--train_sigma_shift",
47
+ "5.0",
48
+ "--use_query_frame_impulse_condition",
49
+ "--use_plucker_camera_control",
50
+ "--enable_v2v_plucker_camera_control",
51
+ "--use_plucker_raymap_tokens",
52
+ "--use_perlayer_camera_control",
53
+ "--camera_injection_mode",
54
+ "projector",
55
+ "--camera_plucker_encoder_mode",
56
+ "patchify",
57
+ "--use_dense_branch",
58
+ "--dense_in_channels",
59
+ "5",
60
+ "--dense_proj_dim",
61
+ "32",
62
+ "--dense_num_residual_blocks",
63
+ "0",
64
+ "--dense_alpha_track",
65
+ "1.0",
66
+ "--dense_conditioning_noise_max_level",
67
+ "0.15",
68
+ "--source_mask_boundary_depth_dropout_prob",
69
+ "0.5",
70
+ "--source_mask_boundary_kernel_size",
71
+ "5",
72
+ "--dense_warp_mode",
73
+ "hybrid_softsplat",
74
+ "--dense_warp_motion_threshold",
75
+ "0.005",
76
+ "--cond_drop_geo_prob",
77
+ "0.1",
78
+ "--cond_drop_track_prob",
79
+ "0.1",
80
+ "--cond_drop_dense_prob",
81
+ "0.0",
82
+ "--initial_step",
83
+ "2530",
84
+ "--output_path",
85
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout",
86
+ "--dataset_num_workers",
87
+ "7",
88
+ "--dataloader_prefetch_factor",
89
+ "2",
90
+ "--drop_text_prob",
91
+ "0.1",
92
+ "--debug_latents_every",
93
+ "500",
94
+ "--allow_existing_checkpoints",
95
+ "--wandb_project",
96
+ "track4d-360",
97
+ "--wandb_name",
98
+ "ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp",
99
+ "--wandb_tags",
100
+ "ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug",
101
+ "--use_wandb",
102
+ "--vxf_init_checkpoint",
103
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors",
104
+ "--use_gradient_checkpointing",
105
+ "--resume_checkpoint",
106
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors",
107
+ "--max_train_steps",
108
+ "2560",
109
+ "--torch_profile_active_steps",
110
+ "5",
111
+ "--torch_profile_wait_steps",
112
+ "3",
113
+ "--torch_profile_warmup_steps",
114
+ "1",
115
+ "--torch_profile_repeat",
116
+ "1",
117
+ "--no_torch_profile_record_shapes",
118
+ "--no_torch_profile_memory"
119
+ ],
120
+ "program": "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/scripts/wan2.1_fun/train_track4d360.py",
121
+ "codePath": "scripts/wan2.1_fun/train_track4d360.py",
122
+ "codePathLocal": "scripts/wan2.1_fun/train_track4d360.py",
123
+ "git": {
124
+ "remote": "git@github.com:NIRVANALAN/track4d-360.git",
125
+ "commit": "b6681ed2127310d14bc5b93ae5f1d20df8b9e907"
126
+ },
127
+ "email": "yushi.lan@outlook.com",
128
+ "root": "/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout",
129
+ "host": "nid010271",
130
+ "executable": "/projects/public/u6ew/yslan_public/yslan_project/miniforge3/envs/th29py312/bin/python3.12",
131
+ "cpu_count": 288,
132
+ "cpu_count_logical": 288,
133
+ "gpu": "NVIDIA GH200 120GB",
134
+ "gpu_count": 4,
135
+ "disk": {
136
+ "/": {
137
+ "total": "255795724288",
138
+ "used": "526450688"
139
+ }
140
+ },
141
+ "memory": {
142
+ "total": "919613407232"
143
+ },
144
+ "gpu_nvidia": [
145
+ {
146
+ "name": "NVIDIA GH200 120GB",
147
+ "memoryTotal": "102625181696",
148
+ "cudaCores": 16896,
149
+ "architecture": "Hopper",
150
+ "uuid": "GPU-c78246be-bb85-931c-a8c3-f605c70f9873"
151
+ },
152
+ {
153
+ "name": "NVIDIA GH200 120GB",
154
+ "memoryTotal": "102625181696",
155
+ "cudaCores": 16896,
156
+ "architecture": "Hopper",
157
+ "uuid": "GPU-d557559f-6b05-d9af-f672-98b04500c99e"
158
+ },
159
+ {
160
+ "name": "NVIDIA GH200 120GB",
161
+ "memoryTotal": "102625181696",
162
+ "cudaCores": 16896,
163
+ "architecture": "Hopper",
164
+ "uuid": "GPU-fc063581-c253-951d-1735-1c26372c8d1e"
165
+ },
166
+ {
167
+ "name": "NVIDIA GH200 120GB",
168
+ "memoryTotal": "102625181696",
169
+ "cudaCores": 16896,
170
+ "architecture": "Hopper",
171
+ "uuid": "GPU-dca24d57-11a6-3aa7-883b-6f639eb27927"
172
+ }
173
+ ],
174
+ "cudaVersion": "12.7",
175
+ "slurm": {
176
+ "cluster_name": "gracehopper",
177
+ "conf": "/var/spool/slurmd/conf-cache/slurm.conf",
178
+ "cpu_bind": "quiet,mask_cpu:0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF",
179
+ "cpu_bind_list": "0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF",
180
+ "cpu_bind_type": "mask_cpu:",
181
+ "cpu_bind_verbose": "quiet",
182
+ "cpus_on_node": "288",
183
+ "distribution": "cyclic",
184
+ "gpus": "32",
185
+ "gpus_on_node": "4",
186
+ "gpus_per_node": "4",
187
+ "gtids": "0",
188
+ "job_account": "brics.u6ew",
189
+ "job_cpus_per_node": "288(x8)",
190
+ "job_end_time": "1780389885",
191
+ "job_gid": "1483804847",
192
+ "job_gpus": "0,1,2,3",
193
+ "job_id": "4982941",
194
+ "job_name": "14b_pair1x49_profile_ablate_denseres0_fsdp2",
195
+ "job_nodelist": "nid[010271,010275,010290,010292,010296-010299]",
196
+ "job_num_nodes": "8",
197
+ "job_partition": "workq",
198
+ "job_qos": "normal",
199
+ "job_start_time": "1780382685",
200
+ "job_uid": "1483804847",
201
+ "job_user": "yushi.u6ew",
202
+ "jobid": "4982941",
203
+ "launch_node_ipaddr": "10.242.18.96",
204
+ "localid": "0",
205
+ "mpi_type": "pmi2",
206
+ "nnodes": "8",
207
+ "nodeid": "0",
208
+ "nodelist": "nid[010271,010275,010290,010292,010296-010299]",
209
+ "nprocs": "8",
210
+ "ntasks": "8",
211
+ "ntasks_per_node": "1",
212
+ "oom_kill_step": "0",
213
+ "prio_process": "0",
214
+ "procid": "0",
215
+ "srun_comm_host": "10.242.18.96",
216
+ "srun_comm_port": "41477",
217
+ "step_gpus": "0,1,2,3",
218
+ "step_id": "0",
219
+ "step_launcher_port": "41477",
220
+ "step_nodelist": "nid[010271,010275,010290,010292,010296-010299]",
221
+ "step_num_nodes": "8",
222
+ "step_num_tasks": "8",
223
+ "step_resv_ports": "26616-26617",
224
+ "step_tasks_per_node": "1(x8)",
225
+ "stepid": "0",
226
+ "submit_dir": "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360",
227
+ "submit_host": "login40",
228
+ "task_pid": "198920",
229
+ "tasks_per_node": "1(x8)",
230
+ "topology_addr": "global.group4.nid010271",
231
+ "topology_addr_pattern": "switch.switch.node",
232
+ "umask": "0022"
233
+ },
234
+ "writerId": "e80r6z0g9k261lnpba92qxhgslcr7ks2"
235
+ }
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/files/wandb-summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"train/track/injector/block_08/weight_norm":71.77714538574219,"train/hbm/post_t5_allocated_gb":28.869243621826172,"train/track/injector/block_34/weight_norm":71.64431762695312,"train/track/injector/block_01/grad_norm":0.0075633530505001545,"train/track/injector/block_20/grad_norm":0.003714467864483595,"train/track/injector/block_03/weight_norm":71.8019027709961,"train/track/injector/block_16/weight_norm":71.75887298583984,"train/loss":0.06805133819580078,"train/timing/step_total_s":12.027139294994413,"train/track/injector/block_24/weight_norm":71.72344207763672,"train/timing/backward_s":5.079326951003168,"train/track/injector/block_02/weight_norm":71.77030944824219,"train/track/injector/block_25/weight_norm":71.73070526123047,"train/track/injector/block_08/grad_norm":0.007247989531606436,"train/timing/unit/WanVideoUnit_TeaCache_s":2.044346729235258e-05,"train/timing/unit/WanVideoUnit_UnifiedSequenceParallel_s":2.825628143909853e-05,"train/track/injector/block_37/weight_norm":71.64723205566406,"train/track/injector/block_27/grad_norm":0.002658056328073144,"train/track/injector/block_32/grad_norm":0.002075374824926257,"train/track/injector/block_12/grad_norm":0.005726680625230074,"train/track/injector/block_30/grad_norm":0.002298362785950303,"_timestamp":1.7803835547502189e+09,"train/track/injector/block_39/grad_norm":0.004338877275586128,"train/track/injector/block_16/grad_norm":0.005405740346759558,"train/track/injector/block_36/grad_norm":0.0018088072538375854,"train/cond_dropout/drop_tracks":0.21875,"train/track/injector/block_03/grad_norm":0.007106975186616182,"train/timing/unit/WanVideoUnit_WarpedAppearance_s":6.011142613715492e-05,"train/track/injector/block_22/grad_norm":0.0046297223307192326,"train/track/injector/weight_norm_mean":71.72119140625,"train/track/injector/block_30/weight_norm":71.7083740234375,"train/hbm/allocated_gb":3.435391426086426,"train/hbm/post_t5_reserved_gb":28.94489097595215,"train/timing/unit/WanVideoUnit_CfgMerger_s":1.2710699593299069e-05,"train/timing/data_wait_s":0.0034507829986978322,"train/timing/optimizer_s":0.03327711300516967,"train/track/injector/block_32/weight_norm":71.66675567626953,"train/track/injector/block_25/grad_norm":0.0028126623947173357,"train/track/injector/block_20/weight_norm":71.7135009765625,"train/track/injector/block_29/weight_norm":71.70845031738281,"train/timing/unit/WanVideoUnit_PromptEmbedder_s":0.00028199722873978317,"train/hbm/reserved_gb":39.55071258544922,"train/track/injector/block_00/grad_norm":0.01542600616812706,"train/track/injector/block_06/weight_norm":71.78461456298828,"train/t5/on_cuda":-1,"train/track/injector/block_04/grad_norm":0.006662918254733086,"train/track/injector/block_18/weight_norm":71.71800231933594,"train/track/injector/block_22/weight_norm":71.71602630615234,"train/track/injector/block_37/grad_norm":0.001781684928573668,"train/timing/other_s":0.6916093489853665,"train/timing/unit/WanVideoUnit_NoiseInitializer_s":0.00017892703181132674,"train/track/injector/block_13/weight_norm":71.73668670654297,"train/track/injector/block_28/weight_norm":71.74348449707031,"train/track/injector/block_07/grad_norm":0.0076438491232693195,"train/track/injector/block_23/weight_norm":71.71672058105469,"train/track/injector/block_15/weight_norm":71.72630310058594,"train/track/injector/block_07/weight_norm":71.78182983398438,"train/track/injector/block_34/grad_norm":0.001947328564710915,"train/track/adapter/weight_norm":17.002622604370117,"train/track/injector/block_10/grad_norm":0.007022695150226355,"train/track/injector/block_09/weight_norm":71.78085327148438,"train/track/injector/block_23/grad_norm":0.003564701182767749,"train/timing/forward_s":6.0979817740008,"train/track/injector/grad_norm_mean":0.004726294428110123,"train/track/injector/block_33/weight_norm":71.6600341796875,"train/timing/microsteps":1,"train/track/injector/block_12/weight_norm":71.75685119628906,"train/track/injector/block_31/grad_norm":0.0020285714417696,"train/track/injector/block_29/grad_norm":0.0028482715133577585,"train/track/injector/block_36/weight_norm":71.64973449707031,"train/track/injector/block_35/weight_norm":71.68326568603516,"train/track/injector/block_10/weight_norm":71.77770233154297,"train/timing/unit/WanVideoUnit_Vista4DStreams_s":5.154164682608098e-05,"train/track/injector/block_14/weight_norm":71.7538833618164,"train/timing/unit/WanVideoUnit_InputVideoEmbedder_XYZCondition_s":1.6001708507537842,"train/track/injector/block_13/grad_norm":0.0054008979350328445,"train/track/injector/block_19/grad_norm":0.00412755599245429,"train/track/injector/block_26/grad_norm":0.0027969893999397755,"train/track/injector/block_02/grad_norm":0.006992134265601635,"train/track/injector/block_21/grad_norm":0.004170014522969723,"train/track/injector/block_38/weight_norm":71.60940551757812,"train/timing/unit/WanVideoUnit_ModeDropoutAugmentation_s":2.9370274205575697e-05,"train/timing/dit_loss_fn_s":2.5807018280029297,"_runtime":518,"train/track/injector/block_05/weight_norm":71.79119110107422,"train/track/injector/weight_norm_max":71.8019027709961,"train/timing/unit/WanVideoUnit_TargetVideoEncoder_s":0.8126809597015381,"train/track/injector/block_04/weight_norm":71.78141784667969,"train/track/injector/block_01/weight_norm":71.73615264892578,"_wandb":{"runtime":518},"train/timing/unit/WanVideoUnit_V2VConcatenate_s":0.00013471435522660613,"train/track/injector/block_06/grad_norm":0.007085887249559164,"train/track/injector/block_28/grad_norm":0.002767539117485285,"train/timing/reduction_s":0.12494410799990874,"train/track/injector/block_39/weight_norm":71.61337280273438,"train/track/injector/block_31/weight_norm":71.70215606689453,"train/timing/unit/WanVideoUnit_ShapeChecker_s":5.647033322020434e-05,"train/timing/unit/WanVideoUnit_CameraControl_NVS_s":0.0032966327853500843,"train/track/injector/block_05/grad_norm":0.007096964865922928,"train/track/injector/block_14/grad_norm":0.005361848510801792,"train/track/injector/block_26/weight_norm":71.73149871826172,"train/track/injector/block_24/grad_norm":0.0032270734664052725,"train/track/injector/block_17/grad_norm":0.0046652089804410934,"_step":2559,"train/track/injector/block_33/grad_norm":0.002168276347219944,"train/track/injector/block_21/weight_norm":71.71273803710938,"train/track/injector/block_11/weight_norm":71.75850677490234,"train/timing/unit/WanVideoUnit_DenseTargetControl_s":0.2419811189174652,"train/track/injector/block_17/weight_norm":71.72594451904297,"train/track/injector/grad_norm_max":0.01542600616812706,"train/track/injector/block_15/grad_norm":0.004815808031708002,"train/timing/unit/WanVideoUnit_TrackPointCondition_s":0.020848633721470833,"train/track/injector/block_19/weight_norm":71.7264633178711,"train/cond_dropout/drop_dense":0.1875,"train/track/injector/block_00/weight_norm":71.62006378173828,"train/epoch":0,"train/track/injector/block_35/grad_norm":0.0018398460233584046,"train/track/injector/block_38/grad_norm":0.002146689221262932,"train/track/injector/block_27/weight_norm":71.7018051147461,"train/track/injector/block_09/grad_norm":0.007202731911092997,"train/track/adapter/grad_norm":0.005933027248829603,"train/track/injector/block_11/grad_norm":0.0059500886127352715,"train/track/injector/block_18/grad_norm":0.004923135042190552,"train/timing/unit/WanVideoUnit_LatentNoiseAugmentation_s":5.086858800495975e-05}
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/logs/debug-internal.log ADDED
@@ -0,0 +1,87 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-06-02T06:50:49.974405605Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-06-02T06:50:49.976821599Z","level":"INFO","msg":"stream: starting","core version":"0.26.0"}
3
+ {"time":"2026-06-02T06:50:50.36107604Z","level":"INFO","msg":"stream: created new stream","id":"nn58xrpu"}
4
+ {"time":"2026-06-02T06:50:50.36114999Z","level":"INFO","msg":"handler: started"}
5
+ {"time":"2026-06-02T06:50:50.362005773Z","level":"INFO","msg":"stream: started"}
6
+ {"time":"2026-06-02T06:50:50.362019596Z","level":"INFO","msg":"writer: started","stream_id":"nn58xrpu"}
7
+ {"time":"2026-06-02T06:50:50.362095786Z","level":"INFO","msg":"sender: started"}
8
+ {"time":"2026-06-02T06:50:51.690481716Z","level":"INFO","msg":"filestream: sending request","total_files":1,"console_offset":0,"console_lines":1}
9
+ {"time":"2026-06-02T06:50:51.888011116Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
10
+ {"time":"2026-06-02T06:51:01.410733865Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":44}
11
+ {"time":"2026-06-02T06:51:01.428514695Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":168}
12
+ {"time":"2026-06-02T06:51:06.691419747Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":0,"events_lines":2,"console_offset":1,"console_lines":122,"uploaded_len":2}
13
+ {"time":"2026-06-02T06:51:06.9238499Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
14
+ {"time":"2026-06-02T06:51:21.69127157Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":2,"events_lines":2}
15
+ {"time":"2026-06-02T06:51:21.913715771Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
16
+ {"time":"2026-06-02T06:51:36.690922534Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":4,"events_lines":2}
17
+ {"time":"2026-06-02T06:51:36.965752742Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
18
+ {"time":"2026-06-02T06:51:51.692183121Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":6,"events_lines":2,"console_offset":123,"console_lines":1}
19
+ {"time":"2026-06-02T06:51:51.980180422Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
20
+ {"time":"2026-06-02T06:52:06.691301178Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":8,"events_lines":2,"console_offset":124,"console_lines":2}
21
+ {"time":"2026-06-02T06:52:07.06257193Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
22
+ {"time":"2026-06-02T06:52:21.69127678Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":0,"history_lines":2,"events_offset":10,"events_lines":2,"console_offset":122,"console_lines":1}
23
+ {"time":"2026-06-02T06:52:22.085867072Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
24
+ {"time":"2026-06-02T06:52:36.69122883Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":2,"history_lines":1,"events_offset":12,"events_lines":2,"console_offset":122,"console_lines":1}
25
+ {"time":"2026-06-02T06:52:36.896820176Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
26
+ {"time":"2026-06-02T06:52:51.691346553Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":3,"history_lines":1,"events_offset":14,"events_lines":2,"console_offset":122,"console_lines":1}
27
+ {"time":"2026-06-02T06:52:51.892989007Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
28
+ {"time":"2026-06-02T06:53:06.691797368Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":4,"history_lines":1,"events_offset":16,"events_lines":2,"console_offset":122,"console_lines":1}
29
+ {"time":"2026-06-02T06:53:06.871983169Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
30
+ {"time":"2026-06-02T06:53:21.691461565Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":5,"history_lines":1,"events_offset":18,"events_lines":2,"console_offset":122,"console_lines":1}
31
+ {"time":"2026-06-02T06:53:21.883845999Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
32
+ {"time":"2026-06-02T06:53:36.693338948Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":6,"history_lines":1,"events_offset":20,"events_lines":2,"console_offset":122,"console_lines":1}
33
+ {"time":"2026-06-02T06:53:36.916559761Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
34
+ {"time":"2026-06-02T06:53:51.69176395Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":7,"history_lines":2,"events_offset":22,"events_lines":2,"console_offset":122,"console_lines":1}
35
+ {"time":"2026-06-02T06:53:51.993609202Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
36
+ {"time":"2026-06-02T06:54:06.690851286Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":24,"events_lines":2,"console_offset":126,"console_lines":1}
37
+ {"time":"2026-06-02T06:54:07.004048439Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
38
+ {"time":"2026-06-02T06:54:21.691620646Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":26,"events_lines":2}
39
+ {"time":"2026-06-02T06:54:22.147645092Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
40
+ {"time":"2026-06-02T06:54:36.690891775Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":28,"events_lines":2}
41
+ {"time":"2026-06-02T06:54:37.065124365Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
42
+ {"time":"2026-06-02T06:54:51.692454398Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":30,"events_lines":2}
43
+ {"time":"2026-06-02T06:54:51.918197063Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
44
+ {"time":"2026-06-02T06:55:06.69053338Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":32,"events_lines":2}
45
+ {"time":"2026-06-02T06:55:06.905887225Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
46
+ {"time":"2026-06-02T06:55:21.691518Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":34,"events_lines":2,"console_offset":122,"console_lines":1}
47
+ {"time":"2026-06-02T06:55:21.947899715Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
48
+ {"time":"2026-06-02T06:55:36.691140706Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":9,"history_lines":1,"events_offset":36,"events_lines":2,"console_offset":122,"console_lines":1}
49
+ {"time":"2026-06-02T06:55:36.952030582Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
50
+ {"time":"2026-06-02T06:55:51.691789048Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":10,"history_lines":2,"events_offset":38,"events_lines":2,"console_offset":122,"console_lines":1}
51
+ {"time":"2026-06-02T06:55:51.898682683Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
52
+ {"time":"2026-06-02T06:56:06.691691779Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":12,"history_lines":1,"events_offset":40,"events_lines":2,"console_offset":122,"console_lines":1}
53
+ {"time":"2026-06-02T06:56:06.985943244Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
54
+ {"time":"2026-06-02T06:56:21.696443399Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":13,"history_lines":1,"events_offset":42,"events_lines":2,"console_offset":122,"console_lines":1}
55
+ {"time":"2026-06-02T06:56:21.906715321Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
56
+ {"time":"2026-06-02T06:56:36.691377746Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":14,"history_lines":2,"events_offset":44,"events_lines":2,"console_offset":122,"console_lines":1}
57
+ {"time":"2026-06-02T06:56:36.931509511Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
58
+ {"time":"2026-06-02T06:56:51.691747719Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":16,"history_lines":1,"events_offset":46,"events_lines":2,"console_offset":122,"console_lines":1}
59
+ {"time":"2026-06-02T06:56:51.960969812Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
60
+ {"time":"2026-06-02T06:57:06.691390544Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":17,"history_lines":1,"events_offset":48,"events_lines":2,"console_offset":122,"console_lines":1}
61
+ {"time":"2026-06-02T06:57:07.050274115Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
62
+ {"time":"2026-06-02T06:57:21.69204273Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":18,"history_lines":2,"events_offset":50,"events_lines":2,"console_offset":122,"console_lines":1}
63
+ {"time":"2026-06-02T06:57:22.186614525Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
64
+ {"time":"2026-06-02T06:57:36.691431427Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":20,"history_lines":1,"events_offset":52,"events_lines":2,"console_offset":122,"console_lines":1}
65
+ {"time":"2026-06-02T06:57:36.925377253Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
66
+ {"time":"2026-06-02T06:57:51.691892104Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":21,"history_lines":1,"events_offset":54,"events_lines":2,"console_offset":122,"console_lines":1}
67
+ {"time":"2026-06-02T06:57:51.935668512Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
68
+ {"time":"2026-06-02T06:58:06.691226293Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":22,"history_lines":2,"events_offset":56,"events_lines":2,"console_offset":122,"console_lines":1}
69
+ {"time":"2026-06-02T06:58:06.90327617Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
70
+ {"time":"2026-06-02T06:58:21.691228117Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":24,"history_lines":1,"events_offset":58,"events_lines":2,"console_offset":122,"console_lines":1}
71
+ {"time":"2026-06-02T06:58:21.922420057Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
72
+ {"time":"2026-06-02T06:58:36.691373952Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":25,"history_lines":1,"events_offset":60,"events_lines":2,"console_offset":122,"console_lines":1}
73
+ {"time":"2026-06-02T06:58:36.932965537Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
74
+ {"time":"2026-06-02T06:58:51.691882754Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":26,"history_lines":2,"events_offset":62,"events_lines":2,"console_offset":122,"console_lines":1}
75
+ {"time":"2026-06-02T06:58:51.913675918Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
76
+ {"time":"2026-06-02T06:59:06.691802803Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":28,"history_lines":1,"events_offset":64,"events_lines":2,"console_offset":122,"console_lines":1}
77
+ {"time":"2026-06-02T06:59:06.9232588Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
78
+ {"time":"2026-06-02T06:59:21.691448365Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":29,"history_lines":1,"events_offset":66,"events_lines":2,"console_offset":122,"console_lines":1}
79
+ {"time":"2026-06-02T06:59:21.938579666Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
80
+ {"time":"2026-06-02T06:59:30.473233165Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
81
+ {"time":"2026-06-02T06:59:30.474469258Z","level":"INFO","msg":"filestream: sending request","total_files":2,"console_offset":127,"console_lines":99,"uploaded_len":3,"complete":true,"exit_code":0}
82
+ {"time":"2026-06-02T06:59:30.683502844Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
83
+ {"time":"2026-06-02T06:59:30.684680347Z","level":"INFO","msg":"handler: operation stats","stats":{}}
84
+ {"time":"2026-06-02T06:59:30.689378199Z","level":"INFO","msg":"stream: closing"}
85
+ {"time":"2026-06-02T06:59:30.689457461Z","level":"INFO","msg":"handler: closed"}
86
+ {"time":"2026-06-02T06:59:30.690169153Z","level":"INFO","msg":"sender: closed"}
87
+ {"time":"2026-06-02T06:59:30.690201056Z","level":"INFO","msg":"stream: closed"}
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/logs/debug.log ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-06-02 06:50:49,349 INFO MainThread:199010 [wandb_setup.py:_flush():81] Current SDK version is 0.26.0
2
+ 2026-06-02 06:50:49,350 INFO MainThread:199010 [wandb_setup.py:_flush():81] Configure stats pid to 199010
3
+ 2026-06-02 06:50:49,350 INFO MainThread:199010 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-06-02 06:50:49,350 INFO MainThread:199010 [wandb_init.py:setup_run_log_directory():721] Logging user logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/logs/debug.log
5
+ 2026-06-02 06:50:49,350 INFO MainThread:199010 [wandb_init.py:setup_run_log_directory():722] Logging internal logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/logs/debug-internal.log
6
+ 2026-06-02 06:50:49,350 INFO MainThread:199010 [wandb_init.py:init():848] calling init triggers
7
+ 2026-06-02 06:50:49,351 INFO MainThread:199010 [wandb_init.py:init():853] wandb.init called with sweep_config: {}
8
+ config: {'learning_rate': 2e-05, 'weight_decay': 0.01, 'num_epochs': 200, 'save_steps': 500, 'num_workers': 7, 'prefetch_factor': 2, 'dataset_base_path': None, 'dataset_metadata_path': None, 'dataset_repeat': 1, 'dataset_num_workers': 7, 'dataloader_prefetch_factor': 2, 'data_file_keys': 'image,video', 'pointmap_stats_path': None, 'model_paths': None, 'model_id_with_origin_paths': 'Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth', 'model_variant': 'vt2v', 'extra_inputs': None, 'track_xyz_normalization': 'minmax', 'track_xyz_quantile_low': 0.02, 'track_xyz_quantile_high': 0.98, 'predict_track_residual': False, 'residual_target_xyz_source': 'clean', 'residual_zero_center_normalization': False, 'residual_linear_normalization': False, 'residual_log_scale_factor': 3.0, 'stream3r_common_scale_stat': 'mean', 'explicit_motion_loss_weight': 0.0, 'explicit_motion_loss_snr_weight': True, 'explicit_motion_loss_snr_sigmoid': True, 'explicit_motion_loss_snr_sigmoid_a': 1.0, 'explicit_motion_loss_snr_sigmoid_b': -5.0, 'explicit_motion_loss_decode_checkpoint': False, 'explicit_motion_loss_decode_chunk_size': 0, 'normalize_track_with_xyz_cond_stats': False, 'prenormalized_track_video_signed_log': False, 'use_xyz_video_condition': False, 'xyz_video_condition_scale': 1.0, 'xyz_video_condition_aug_prob': 0.0, 'xyz_video_condition_noise_std': 0.0, 'xyz_video_condition_z_noise_std': 0.0, 'xyz_video_condition_frame_scale_std': 0.0, 'xyz_video_condition_dropout_prob': 0.0, 'xyz_video_condition_mode': 'xyz', 'xyz_video_condition_cfg_dropout_prob': 0.0, 'camera_condition_cfg_dropout_prob': 0.0, 'fp8_models': None, 'offload_models': None, 'v2v_concat_mode': 'width', 'v2v_channel_patch_embed_mode': 'add', 'trainable_models': 'dit.domain_embedding,dit.control_patch_embedding', 'find_unused_parameters': False, 'task': 'sft:v2v', 'drop_text_prob': 0.1, 'drop_video_prob': 0.0, 'drop_t2v_prob': None, 'drop_i2v_prob': None, 'input_video_noise_strength': 0.0, 'input_video_noise_random_max': 500, 'output_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout', 'remove_prefix_in_ckpt': 'pipe.dit.', 'save_every_n_epochs': 2, 'max_checkpoints_to_keep': 5, 'allow_existing_checkpoints': True, 'initial_step': 2530, 'max_train_steps': 2560, 'lora_base_model': None, 'lora_target_modules': 'q,k,v,o,ffn.0,ffn.2', 'lora_rank': 32, 'lora_alpha': None, 'lora_checkpoint': None, 'preset_lora_path': None, 'preset_lora_model': None, 'use_gradient_checkpointing': True, 'use_gradient_checkpointing_offload': False, 'gradient_accumulation_steps': 1, 'use_wandb': True, 'wandb_project': 'track4d-360', 'wandb_name': 'ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp', 'wandb_entity': None, 'wandb_tags': ['ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug'], 'wandb_notes': None, 'wandb_log_gradients': False, 'wandb_log_parameters': False, 'wandb_log_freq': 100, 'wandb_watch_model': False, 'debug_latents': False, 'debug_latents_every': 500, 'debug_latents_fps': 16, 'debug_latents_quality': 5, 'debug_latents_max_frames': None, 'debug_latents_outdir': 'debug_latents', 'debug_latents_vae_tiled': True, 'debug_latents_vae_tile_size': (30, 52), 'debug_latents_vae_tile_stride': (15, 26), 'debug_missing_grads': False, 'torch_distributed_debug': None, 'use_query_frame_impulse_condition': True, 'height': 288, 'width': 512, 'max_pixels': 1048576, 'num_frames': 49, 'tokenizer_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl', 'dit_finetune_modules': 'self_attn', 'track_injection_mode': 'per_block', 'track_injection_block_mode': 'concat_project', 'train_sigma_shift': 5.0, 'dummy': False, 'steps_per_epoch': 25000, 'initialize_model_on_cpu': False, 'resume_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors', 'gs_head_init_checkpoint': None, 'joint_track_prediction': False, 'joint_track_depth_prediction': False, 'joint_depth_prediction': False, 'target_depth_log_cap': 14.0, 'target_depth_norm_type': 'log_depth', 'target_depth_max_disparity': 5.0, 'vxf_init_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors', 'channel_concat_stage': None, 'use_residual_tracks': True, 'freeze_lora_after_load': False, 'skip_v2v_required_trainables': False, 'gs_only_training': False, 'freeze_track_block_injector': False, 'freeze_track4d_condition_modules': False, 'gs_head_style': 'flashworld', 'use_gs_output': False, 'gs_zero_features': False, 'gs_feat_dim': 1024, 'gs_loss_weight': 1.0, 'gs_mse_loss_weight': 1.0, 'gs_lpips_loss_weight': 0.5, 'gs_depth_loss_weight': 0.05, 'gs_abs_depth_loss_weight': 0.0, 'gs_opacity_loss_weight': 0.1, 'gs_opacity_prune_keep_frac': 1.0, 'gs_rgb_loss_weight': None, 'gs_sh_degree': 2, 'gs_num_points_per_pixel': 2, 'no_gs_latent_detach': False, 'use_gs_feedback_training': False, 'gs_scale_cap': None, 'gs_render_patch_size': None, 'gs_legacy_union_render': False, 'gs_lyra_depth_param': 'exp', 'gs_lyra_log_depth_shift': 0.0, 'gs_lyra_dnear': 0.1, 'gs_lyra_dfar': 10.0, 'gs_replay_cache_size': 0, 'gs_dynamic_mode': 'static', 'gs_track_supervision_mode': 'pointmap', 'gs_pos_loss_weight': 1.0, 'gs_dyn_depth_loss_weight': 0.5, 'gs_opa_tv_weight': 0.01, 'gs_xyz_tv_weight': 0.01, 'gs_predict_rotation': False, 'gs_disable_delta_opacity': False, 'gs_direct_pos_loss_weight': 0.0, 'gs_track_canonical_loss_weight': 0.0, 'gs_track_delta_loss_weight': 0.0, 'gs_track_final_pos_loss_weight': 0.0, 'gs_dynamic_max_t_video': 64, 'gs_assoc_track_2d_loss_weight': 0.0, 'gs_track_accel_loss_weight': 0.0, 'gs_track_loss_quantile': 0.98, 'gs_track_canonical_detach': False, 'gs_track_joint_render': False, 'cond_drop_track_prob': 0.1, 'cond_drop_dense_prob': 0.0, 'cond_drop_geo_prob': 0.1, 'use_track_null_token_fallback': False, 'use_dense_null_token_fallback': False, 'use_plucker_camera_control': True, 'enable_v2v_plucker_camera_control': True, 'train_plucker_camera_control': False, 'use_plucker_raymap_tokens': True, 'plucker_use_ray_origin': False, 'use_perlayer_camera_control': True, 'camera_injection_mode': 'projector', 'camera_plucker_encoder_mode': 'patchify', 'warped_condition_mode': 'off', 'warped_appearance_fusion': 'concat_proj', 'warped_geom_only_dense': True, 'multi_frame_warped_appearance': False, 'multi_frame_motion_threshold': 0.01, 'fixed_step_frac': None, 'fixed_num_inference_steps': 50, 'fixed_sigma': None, 'skip_diffusion_loss': False, 'use_dense_branch': True, 'dense_in_channels': 5, 'dense_proj_dim': 32, 'dense_num_residual_blocks': 0, 'dense_alpha_track': 1.0, 'use_depth_null_token': False, 'dense_conditioning_noise_max_level': 0.15, 'source_mask_boundary_depth_dropout_prob': 0.5, 'source_mask_boundary_kernel_size': 5, 'dense_warp_mode': 'hybrid_softsplat', 'dense_warp_motion_threshold': 0.005, 'dense_hybrid_disable_dynamic_masks': False, 'use_vista4d_latent_streams': False, 'use_gen3c_warp_latent_channels': False, 'disable_track_modules': False, 'dense_only_condition_modules': False, 'torch_profile_active_steps': 5, 'torch_profile_wait_steps': 3, 'torch_profile_warmup_steps': 1, 'torch_profile_repeat': 1, 'torch_profile_dir': None, 'torch_profile_record_shapes': False, 'torch_profile_memory': False, 'torch_profile_with_stack': False, 'track_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth.yaml', 'track4d360_data_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml', 'syn4d_root': '/scratch/shared/beegfs/zeren/Syn4D', 'scene': None, 'sequence_idx': None, 'track_spatial_stride': 1, '_wandb': {}}
9
+ 2026-06-02 06:50:49,351 INFO MainThread:199010 [wandb_init.py:init():896] starting backend
10
+ 2026-06-02 06:50:49,963 INFO MainThread:199010 [wandb_init.py:init():911] sending inform_init request
11
+ 2026-06-02 06:50:49,972 INFO MainThread:199010 [wandb_init.py:init():919] backend started and connected
12
+ 2026-06-02 06:50:49,975 INFO MainThread:199010 [wandb_init.py:init():989] updated telemetry
13
+ 2026-06-02 06:50:50,021 INFO MainThread:199010 [wandb_init.py:init():1013] communicating run to backend with 90.0 second timeout
14
+ 2026-06-02 06:50:50,887 INFO MainThread:199010 [wandb_init.py:init():1058] starting run threads in backend
15
+ 2026-06-02 06:50:51,673 INFO MainThread:199010 [wandb_run.py:_console_start():2542] atexit reg
16
+ 2026-06-02 06:50:51,674 INFO MainThread:199010 [wandb_run.py:_redirect():2391] redirect: wrap_raw
17
+ 2026-06-02 06:50:51,674 INFO MainThread:199010 [wandb_run.py:_redirect():2460] Wrapping output streams.
18
+ 2026-06-02 06:50:51,674 INFO MainThread:199010 [wandb_run.py:_redirect():2483] Redirects installed.
19
+ 2026-06-02 06:50:51,682 INFO MainThread:199010 [wandb_init.py:init():1098] run started, returning control to user process
20
+ 2026-06-02 06:59:29,656 INFO MainThread:199010 [wandb_run.py:_finish():2309] finishing run nirvanalan/track4d-360/nn58xrpu
21
+ 2026-06-02 06:59:29,663 INFO MainThread:199010 [wandb_run.py:_atexit_cleanup():2508] got exitcode: 0
22
+ 2026-06-02 06:59:29,664 INFO MainThread:199010 [wandb_run.py:_restore():2490] restore
23
+ 2026-06-02 06:59:29,664 INFO MainThread:199010 [wandb_run.py:_restore():2496] restore done
24
+ 2026-06-02 06:59:30,688 INFO MainThread:199010 [wandb_run.py:_footer_sync_info():3886] logging synced files
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_ablate_denseres0_cond_dropout/wandb/run-20260602_065049-nn58xrpu/run-nn58xrpu.wandb ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:019ed6c53cc75fb8ede1e41b3a1f83dae28c798b81f4840269386ee945160586
3
+ size 674396
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/ismb_sbatch_14b_8node_288x512_fsdp_raymaptokens_nextstage_recammaster_pair1x49_eb32_lr2e5.sh ADDED
@@ -0,0 +1,170 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+ #SBATCH --job-name=14b_8node_288x512_pair1x49_next_eb32_lr2e5
3
+ #SBATCH --nodes=8
4
+ #SBATCH --gpus=32
5
+ #SBATCH --gpus-per-node=4
6
+ #SBATCH --ntasks-per-node=1
7
+ #SBATCH --time=24:00:00
8
+ #SBATCH --output=out/exp_%j_%x.out
9
+
10
+ # Track4D-360 14B FSDP 288x512 next-stage finetune.
11
+ #
12
+ # Main changes from the 288p backendmix OOM-debug launcher:
13
+ # - new DATA_TAG / output directory;
14
+ # - warm-start from the latest existing 288p backendmix checkpoint;
15
+ # - remove most static datasets from DATA_CONFIG;
16
+ # - add ReCamMaster CoWTracker + square512 source49 + compact pair1x49
17
+ # VGGT-Omega labels, q1-only by dataset construction.
18
+ #
19
+ # Submit from repo root on Isambard:
20
+ # sbatch bash_scripts/track4d_360/ismb/14b/sbatch/ismb_sbatch_14b_8node_288x512_fsdp_raymaptokens_nextstage_recammaster_pair1x49_eb32_lr2e5.sh
21
+
22
+ export NNODES=8
23
+ export MODEL_SIZE=14b
24
+ export CONDA_ENV="${CONDA_ENV:-th29py312}"
25
+ export HEIGHT=288
26
+ export WIDTH=512
27
+ export NUM_FRAMES=49
28
+ export main_process_port="${main_process_port:-29595}"
29
+
30
+ export WAN_T2V14B_DIR="${WAN_T2V14B_DIR:-${SLURM_SUBMIT_DIR}/weights/Wan2.1-T2V-14B}"
31
+ export VXF_INIT="${VXF_INIT:-${WAN_T2V14B_DIR}/diffusion_pytorch_model*.safetensors}"
32
+ export MODEL_ID_PREFIX="${MODEL_ID_PREFIX:-Wan-AI/Wan2.1-T2V-14B}"
33
+ export TOKENIZER_PATH="${TOKENIZER_PATH:-${WAN_T2V14B_DIR}/google/umt5-xxl}"
34
+ export T5_PATH="${T5_PATH:-${WAN_T2V14B_DIR}/models_t5_umt5-xxl-enc-bf16.pth}"
35
+ export VAE_PATH="${VAE_PATH:-${WAN_T2V14B_DIR}/Wan2.1_VAE.pth}"
36
+
37
+ # Prefer the current 288p backendmix run as the next-stage warm start. Fall
38
+ # back to the 384p backendmix/raymap-token parents only when no 288p checkpoint exists.
39
+ export BACKENDMIX_288_PARENT_DIR="${BACKENDMIX_288_PARENT_DIR:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_warmstart384_backendmix_q1to4_s1248_jitter_norecam_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_warmstart384_backendmix_q1to4_s1248_jitter_norecam_eb32_288x512_cond_dropout}"
40
+ export BACKENDMIX_384_PARENT_DIR="${BACKENDMIX_384_PARENT_DIR:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_resume_latest_backendmix_q1to4_s1248_jitter_norecam_eb32_384x672_cond_dropout_384x672x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_real_synth_syn4d_softsplat_raymaptokens_resume_latest_backendmix_q1to4_s1248_jitter_norecam_eb32_384x672_cond_dropout}"
41
+ export RAYMAP_384_PARENT_DIR="${RAYMAP_384_PARENT_DIR:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_real_synth_syn4d_recam_syncam_softsplat_raymaptokens_t2v_init_scratch_eb32_cond_dropout_384x672x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_real_synth_syn4d_recam_syncam_softsplat_raymaptokens_t2v_init_scratch_eb32_cond_dropout}"
42
+ # Warm-start from the 288p backendmix run's step-9000 checkpoint — the requested
43
+ # next-stage starting point. (FSDP1/FSDP2 read it identically: warm-start loads
44
+ # the safetensors BEFORE accelerator.prepare, on the unwrapped model, so it is
45
+ # distributed-backend-independent.) Override WARM_START=... to pick another;
46
+ # falls back to latest backendmix / 384p parents if step-9000 is absent.
47
+ if [ -z "${WARM_START:-}" ]; then
48
+ if [ -f "${BACKENDMIX_288_PARENT_DIR}/step-9000.safetensors" ]; then
49
+ export WARM_START="${BACKENDMIX_288_PARENT_DIR}/step-9000.safetensors"
50
+ else
51
+ latest="$(find "${BACKENDMIX_288_PARENT_DIR}" -maxdepth 1 -name 'step-*.safetensors' 2>/dev/null | sort -V | tail -1)"
52
+ if [ -z "${latest}" ]; then
53
+ latest="$(find "${BACKENDMIX_384_PARENT_DIR}" -maxdepth 1 -name 'step-*.safetensors' 2>/dev/null | sort -V | tail -1)"
54
+ fi
55
+ if [ -z "${latest}" ]; then
56
+ latest="$(find "${RAYMAP_384_PARENT_DIR}" -maxdepth 1 -name 'step-*.safetensors' 2>/dev/null | sort -V | tail -1)"
57
+ fi
58
+ export WARM_START="${latest:-${RAYMAP_384_PARENT_DIR}/step-3500.safetensors}"
59
+ fi
60
+ fi
61
+
62
+ if ! compgen -G "${VXF_INIT}" >/dev/null; then
63
+ echo "ERROR: VXF_INIT not found: ${VXF_INIT}" >&2
64
+ exit 1
65
+ fi
66
+ [ -d "${TOKENIZER_PATH}" ] || { echo "ERROR: TOKENIZER_PATH not found: ${TOKENIZER_PATH}" >&2; exit 1; }
67
+ [ -f "${T5_PATH}" ] || { echo "ERROR: T5_PATH not found: ${T5_PATH}" >&2; exit 1; }
68
+ [ -f "${VAE_PATH}" ] || { echo "ERROR: VAE_PATH not found: ${VAE_PATH}" >&2; exit 1; }
69
+
70
+ export TARGET_EFFECTIVE_BATCH=32
71
+ export gradient_accumulation_steps=1
72
+ export learning_rate="${learning_rate:-2e-5}"
73
+ export save_steps="${save_steps:-500}"
74
+ export dataset_num_workers="${dataset_num_workers:-7}"
75
+ export TRACK4D360_STOP_AFTER_NEXT_SAVE="${TRACK4D360_STOP_AFTER_NEXT_SAVE:-0}"
76
+
77
+ export dit_finetune_modules="${dit_finetune_modules:-self_attn}"
78
+ export DIT_FINETUNE_TAG="${DIT_FINETUNE_TAG:-attn}"
79
+ export use_plucker_raymap_tokens=1
80
+ export train_plucker_camera_control=0
81
+ export PLUCKER_WANDB_TAG="${PLUCKER_WANDB_TAG:-plucker_raymap_tokens}"
82
+
83
+ export DATA_CONFIG="${DATA_CONFIG:-${SLURM_SUBMIT_DIR}/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml}"
84
+ export DATA_TAG="${DATA_TAG:-mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512}"
85
+
86
+ export RECAMMASTER_SOURCE49_ROOT="${RECAMMASTER_SOURCE49_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_square512_source49}"
87
+ export RECAMMASTER_PAIR1X49_LABEL_ROOT="${RECAMMASTER_PAIR1X49_LABEL_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_pair1x49_square512_scale_labels}"
88
+ export RECAMMASTER_COW_ROOT="${RECAMMASTER_COW_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_VGGT_Omega_CoWTracker_SourceTracks_all}"
89
+ export TRACK4D360_RECAM_TRACK_SAMPLE_MODE="${TRACK4D360_RECAM_TRACK_SAMPLE_MODE:-grid}"
90
+ export TRACK4D360_RECAM_TRACK_STRIDE="${TRACK4D360_RECAM_TRACK_STRIDE:-2}"
91
+ export TRACK4D360_RECAM_TRACK_SAMPLES="${TRACK4D360_RECAM_TRACK_SAMPLES:-0}"
92
+ # Pair1x49 labels currently cover only the priority block: first 750 numeric
93
+ # train scenes inside each K/focal group (e.g. train/f18_aperture10/scene000..).
94
+ export TRACK4D360_RECAM_SCENE_OFFSET_PER_K="${TRACK4D360_RECAM_SCENE_OFFSET_PER_K:-0}"
95
+ export TRACK4D360_RECAM_SCENE_LIMIT_PER_K="${TRACK4D360_RECAM_SCENE_LIMIT_PER_K:-750}"
96
+
97
+ # Fixed post-compaction track identity count — MUST be set (the 24h/10k-step
98
+ # backendmix run had this at 65536; this launcher previously omitted it). The
99
+ # ReCamMaster CoW tracks are dense (grid stride 2, samples=0 = uncapped), so
100
+ # without a cap N varies per sample and a heavy draw spikes the track adapter's
101
+ # proj_down activation (B,N,T,hidden) -> the observed CUDA OOM at
102
+ # conv3d_track_adapter.py proj_down (1.16 GiB alloc, step ~770). Fixing N also
103
+ # pads small draws to a constant shape, eliminating per-step fragmentation. Lower
104
+ # to 32768 for extra headroom on contended nodes.
105
+ export TRACK4D360_TRACK_IDENTITY_FIXED_N="${TRACK4D360_TRACK_IDENTITY_FIXED_N:-65536}"
106
+
107
+ export TRACK4D360_SPARSE_TRACK_PROB="${TRACK4D360_SPARSE_TRACK_PROB:-1.0}"
108
+ export TRACK4D360_SPARSE_TRACK_STRIDE="${TRACK4D360_SPARSE_TRACK_STRIDE:-2|4|8}"
109
+ export TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB="${TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB:-1.0}"
110
+ export TRACK4D360_SPARSE_TRACK_BLOCK_SIZE="${TRACK4D360_SPARSE_TRACK_BLOCK_SIZE:-64}"
111
+ export TRACK4D360_MULTI_QUERY_PROB="${TRACK4D360_MULTI_QUERY_PROB:-1.0}"
112
+ export TRACK4D360_MULTI_QUERY_CHOICES="${TRACK4D360_MULTI_QUERY_CHOICES:-1|2|3|4}"
113
+ export TRACK4D360_GEOMETRY_NOISE_PROB="${TRACK4D360_GEOMETRY_NOISE_PROB:-1.0}"
114
+ export TRACK4D360_TRACK_OFFSET_STD="${TRACK4D360_TRACK_OFFSET_STD:-0.015}"
115
+ export TRACK4D360_TRACK_TEMPORAL_JITTER_STD="${TRACK4D360_TRACK_TEMPORAL_JITTER_STD:-0.006}"
116
+ export TRACK4D360_TRACK_XY_NOISE_RATIO="${TRACK4D360_TRACK_XY_NOISE_RATIO:-0.25}"
117
+ export TRACK4D360_TRACK_FRAME_DEPTH_JITTER_RATIO="${TRACK4D360_TRACK_FRAME_DEPTH_JITTER_RATIO:-0.5}"
118
+ export TRACK4D360_DEPTH_OFFSET_STD="${TRACK4D360_DEPTH_OFFSET_STD:-0.006}"
119
+ export TRACK4D360_DEPTH_SCALE_JITTER="${TRACK4D360_DEPTH_SCALE_JITTER:-0.01}"
120
+ export TRACK4D360_SOURCE_RGB_NOISE_STD="${TRACK4D360_SOURCE_RGB_NOISE_STD:-0.01}"
121
+ export TRACK4D360_SOURCE_RGB_FRAME_JITTER_STD="${TRACK4D360_SOURCE_RGB_FRAME_JITTER_STD:-0.02}"
122
+ export TRACK4D360_DYNAMIC_MASK_AUG_PROB="${TRACK4D360_DYNAMIC_MASK_AUG_PROB:-0.5}"
123
+ export TRACK4D360_DYNAMIC_MASK_MORPH_KERNEL="${TRACK4D360_DYNAMIC_MASK_MORPH_KERNEL:-5}"
124
+ export TRACK4D360_DYNAMIC_MASK_FLIP_PROB="${TRACK4D360_DYNAMIC_MASK_FLIP_PROB:-0.002}"
125
+
126
+ export dense_warp_mode="${dense_warp_mode:-hybrid_softsplat}"
127
+ export dense_warp_motion_threshold="${dense_warp_motion_threshold:-0.005}"
128
+ export dense_conditioning_noise_max_level="${dense_conditioning_noise_max_level:-0.15}"
129
+ export source_mask_boundary_depth_dropout_prob="${source_mask_boundary_depth_dropout_prob:-0.5}"
130
+ export source_mask_boundary_kernel_size="${source_mask_boundary_kernel_size:-5}"
131
+ export freeze_track4d_condition_modules="${freeze_track4d_condition_modules:-0}"
132
+
133
+ export use_gradient_checkpointing=1
134
+ export use_gradient_checkpointing_offload="${use_gradient_checkpointing_offload:-0}"
135
+ export VXF_OFFLOAD_T5_TO_CPU="${VXF_OFFLOAD_T5_TO_CPU:-1}"
136
+ export TRACK4D360_MEM_PROBE="${TRACK4D360_MEM_PROBE:-0}"
137
+ export TRACK4D360_ADAPTER_MEM_PROFILE="${TRACK4D360_ADAPTER_MEM_PROFILE:-0}"
138
+ export PYTORCH_CUDA_ALLOC_CONF="${PYTORCH_CUDA_ALLOC_CONF:-${PYTORCH_ALLOC_CONF:-expandable_segments:True}}"
139
+ export PYTORCH_ALLOC_CONF="${PYTORCH_ALLOC_CONF:-${PYTORCH_CUDA_ALLOC_CONF}}"
140
+ export TRACK4D360_DENSE_HYBRID_CACHE_FRAMES="${TRACK4D360_DENSE_HYBRID_CACHE_FRAMES:-7}"
141
+ export TRACK4D360_DENSE_HYBRID_TGT_CHUNK="${TRACK4D360_DENSE_HYBRID_TGT_CHUNK:-2}"
142
+
143
+ export NCCL_P2P_DISABLE="${NCCL_P2P_DISABLE:-0}"
144
+ export NCCL_P2P_LEVEL="${NCCL_P2P_LEVEL:-NVL}"
145
+ export NCCL_NVLS_ENABLE=0
146
+ export NCCL_NSOCKS_PERTHREAD=8
147
+ export NCCL_SOCKET_NTHREADS=8
148
+ export NCCL_MAX_NCHANNELS="${NCCL_MAX_NCHANNELS:-32}"
149
+ export NCCL_BUFFSIZE="${NCCL_BUFFSIZE:-8388608}"
150
+ export NCCL_DEBUG="${NCCL_DEBUG:-WARN}"
151
+
152
+ export TRACK4D360_CAPTION_OUTPUT_ROOT="${TRACK4D360_CAPTION_OUTPUT_ROOT:-/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output}"
153
+ export SYN4D_CAPTION_DIR="${SYN4D_CAPTION_DIR:-${TRACK4D360_CAPTION_OUTPUT_ROOT}/Syn4D}"
154
+
155
+ echo "[raymap-tokens 14B pair1x49 next] T2V init: ${VXF_INIT}"
156
+ echo "[raymap-tokens 14B pair1x49 next] warm start: ${WARM_START}"
157
+ echo "[raymap-tokens 14B pair1x49 next] data: ${DATA_CONFIG}"
158
+ echo "[raymap-tokens 14B pair1x49 next] ReCam source49: ${RECAMMASTER_SOURCE49_ROOT}"
159
+ echo "[raymap-tokens 14B pair1x49 next] ReCam pair1x49 labels: ${RECAMMASTER_PAIR1X49_LABEL_ROOT}"
160
+ echo "[raymap-tokens 14B pair1x49 next] ReCam CoW: ${RECAMMASTER_COW_ROOT}"
161
+ echo "[raymap-tokens 14B pair1x49 next] ReCam q1 track sampling: mode=${TRACK4D360_RECAM_TRACK_SAMPLE_MODE}, stride=${TRACK4D360_RECAM_TRACK_STRIDE}, samples=${TRACK4D360_RECAM_TRACK_SAMPLES}"
162
+ echo "[raymap-tokens 14B pair1x49 next] ReCam scene window per K: offset=${TRACK4D360_RECAM_SCENE_OFFSET_PER_K}, limit=${TRACK4D360_RECAM_SCENE_LIMIT_PER_K}"
163
+ echo "[raymap-tokens 14B pair1x49 next] q policy: ReCam q1-only; synthetic choices=${TRACK4D360_MULTI_QUERY_CHOICES}"
164
+ echo "[raymap-tokens 14B pair1x49 next] sparse synthetic tracks: stride_choices=${TRACK4D360_SPARSE_TRACK_STRIDE}, block_jitter_prob=${TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB}"
165
+ echo "[raymap-tokens 14B pair1x49 next] trainables: self_attn + raymap Ecam/camera_projector + active track/dense condition modules"
166
+ echo "[raymap-tokens 14B pair1x49 next] freeze track/dense condition modules: ${freeze_track4d_condition_modules}"
167
+ echo "[raymap-tokens 14B pair1x49 next] batch: effective=32, grad_accum=1, lr=${learning_rate}, save_steps=${save_steps}"
168
+ echo "[raymap-tokens 14B pair1x49 next] memory: grad_ckpt_offload=${use_gradient_checkpointing_offload}, hybrid_cache_frames=${TRACK4D360_DENSE_HYBRID_CACHE_FRAMES}, hybrid_tgt_chunk=${TRACK4D360_DENSE_HYBRID_TGT_CHUNK}, cuda_alloc_conf=${PYTORCH_CUDA_ALLOC_CONF}"
169
+
170
+ source "${SLURM_SUBMIT_DIR}/bash_scripts/track4d_360/ismb/sbatch_common.sh"
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Next-stage 288x512 backendmix finetune with cleaned ReCamMaster pair1x49 labels.
2
+ #
3
+ # Mix: 35% ReCamMaster pair1x49, 35% Syn4D, 5% Kubric49, 5% Kubric81,
4
+ # and 20% legacy static datasets. The static 20% keeps the previous backendmix
5
+ # relative weights: DL3DV:RE10K:MVS-Synth:TartanAir = 65000:7500:12500:25000.
6
+ # ReCamMaster uses compact square512 source49 + pair1x49 VGGT-Omega scale
7
+ # labels plus CoWTracker source tracks and is q1-only by construction.
8
+ #
9
+ # sbatch_common.sh still sanity-checks for the old MVS-Synth guard string:
10
+ # exclude_sequence_ids=['0000']. This config has no MVS-Synth entry.
11
+
12
+ num_views: 49
13
+ resolution: [512, 288]
14
+
15
+ syn4d_root: ${oc.env:SYN4D_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/Syn4D}
16
+ syn4d_caption_dir: ${oc.env:SYN4D_CAPTION_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/Syn4D}
17
+ syn4d_valid_sequences: configs/syn4d_valid_sequences.txt
18
+ dl3dv_root: ${oc.env:DL3DV_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/dl3dv_tar}
19
+ re10k_root: ${oc.env:RE10K_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/processed_re10k}
20
+ mvs_synth_root: ${oc.env:MVS_SYNTH_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/processed_mvs_synth}
21
+ tartanair_root: ${oc.env:TARTANAIR_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/cut3r_processed/processed_tartanair}
22
+ kubric49_root: ${oc.env:KUBRIC49_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/track_dataset/train_preprocessed_49frames}
23
+ kubric81_root: ${oc.env:KUBRIC81_ROOT,/lus/lfs1aip2/projects/public/u6cl/yslan_data/track_dataset/train_preprocessed_81frames}
24
+ kubric49_caption_cache_dir: ${oc.env:KUBRIC49_CAPTION_CACHE_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/kubric_train_49}
25
+ kubric81_caption_cache_dir: ${oc.env:KUBRIC81_CAPTION_CACHE_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/kubric_train_81}
26
+ recammaster_caption_cache_dir: ${oc.env:RECAMMASTER_CAPTION_CACHE_DIR,/lus/lfs1aip2/projects/public/u6ew/yslan_data/caption_output/recammaster}
27
+ recammaster_root: ${oc.env:RECAMMASTER_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset}
28
+ multicam_fg_mask_root: ${oc.env:MULTICAM_FG_MASK_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_SAM3_FG_81f_sources_person_human_288x512}
29
+ recammaster_source49_root: ${oc.env:RECAMMASTER_SOURCE49_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_square512_source49}
30
+ recammaster_pair1x49_label_root: ${oc.env:RECAMMASTER_PAIR1X49_LABEL_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_4RC_VGGT_Omega_pair1x49_square512_scale_labels}
31
+ recammaster_cow_root: ${oc.env:RECAMMASTER_COW_ROOT,/lus/lfs1aip2/projects/public/u6ew/yslan_public/dataset/MultiCamVideo-Dataset_VGGT_Omega_CoWTracker_SourceTracks_all}
32
+ recammaster_track_sample_mode: ${oc.env:TRACK4D360_RECAM_TRACK_SAMPLE_MODE,grid}
33
+ recammaster_track_stride: ${oc.env:TRACK4D360_RECAM_TRACK_STRIDE,2}
34
+ recammaster_track_samples: ${oc.env:TRACK4D360_RECAM_TRACK_SAMPLES,0}
35
+ recammaster_scene_offset_per_k: ${oc.env:TRACK4D360_RECAM_SCENE_OFFSET_PER_K,0}
36
+ recammaster_scene_limit_per_k: ${oc.env:TRACK4D360_RECAM_SCENE_LIMIT_PER_K,750}
37
+
38
+ sparse_track_prob: ${oc.env:TRACK4D360_SPARSE_TRACK_PROB,1.0}
39
+ sparse_track_stride: ${oc.env:TRACK4D360_SPARSE_TRACK_STRIDE,2|4|8}
40
+ sparse_track_query_stride_mode: ${oc.env:TRACK4D360_SPARSE_TRACK_QUERY_STRIDE_MODE,query_count_floor}
41
+ sparse_track_block_jitter_prob: ${oc.env:TRACK4D360_SPARSE_TRACK_BLOCK_JITTER_PROB,1.0}
42
+ sparse_track_block_size: ${oc.env:TRACK4D360_SPARSE_TRACK_BLOCK_SIZE,64}
43
+ multi_query_prob: ${oc.env:TRACK4D360_MULTI_QUERY_PROB,1.0}
44
+ multi_query_choices: ${oc.env:TRACK4D360_MULTI_QUERY_CHOICES,1|2|3|4}
45
+ geometry_noise_prob: ${oc.env:TRACK4D360_GEOMETRY_NOISE_PROB,1.0}
46
+ track_offset_std: ${oc.env:TRACK4D360_TRACK_OFFSET_STD,0.015}
47
+ track_temporal_jitter_std: ${oc.env:TRACK4D360_TRACK_TEMPORAL_JITTER_STD,0.006}
48
+ track_xy_noise_ratio: ${oc.env:TRACK4D360_TRACK_XY_NOISE_RATIO,0.25}
49
+ track_frame_depth_jitter_ratio: ${oc.env:TRACK4D360_TRACK_FRAME_DEPTH_JITTER_RATIO,0.5}
50
+ depth_offset_std: ${oc.env:TRACK4D360_DEPTH_OFFSET_STD,0.006}
51
+ depth_scale_jitter: ${oc.env:TRACK4D360_DEPTH_SCALE_JITTER,0.01}
52
+ source_rgb_noise_std: ${oc.env:TRACK4D360_SOURCE_RGB_NOISE_STD,0.01}
53
+ source_rgb_frame_jitter_std: ${oc.env:TRACK4D360_SOURCE_RGB_FRAME_JITTER_STD,0.02}
54
+ dynamic_mask_aug_prob: ${oc.env:TRACK4D360_DYNAMIC_MASK_AUG_PROB,0.5}
55
+ dynamic_mask_morph_kernel: ${oc.env:TRACK4D360_DYNAMIC_MASK_MORPH_KERNEL,5}
56
+ dynamic_mask_flip_prob: ${oc.env:TRACK4D360_DYNAMIC_MASK_FLIP_PROB,0.002}
57
+
58
+ train_datasets:
59
+ # 35% ReCamMaster CoWTracker + square512 source49 + compact pair1x49 VGGT labels.
60
+ # This dataset is q1 only: it always emits track_query_indices=[0]. The
61
+ # default track density is grid stride 2 to avoid lifting every CoW pixel
62
+ # during training; override TRACK4D360_RECAM_TRACK_SAMPLE_MODE=all for dense.
63
+ - "87500 @ ReCamMasterCowVGGT_NVS(ROOT='${recammaster_root}', vggt_root='${recammaster_source49_root}', source_vggt_root='${recammaster_source49_root}', pair1x49_label_root='${recammaster_pair1x49_label_root}', geometry_source='pair1x49_label_scale_only', cow_root='${recammaster_cow_root}', fg_mask_root='${multicam_fg_mask_root}', caption_cache_dir='${recammaster_caption_cache_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=87500, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, vggt_post_crop_mode='cow_aspect', rgb_post_crop_mode='cow_aspect', track_sample_mode='${recammaster_track_sample_mode}', track_stride=${recammaster_track_stride}, track_samples=${recammaster_track_samples}, scene_offset_per_k=${recammaster_scene_offset_per_k}, scene_limit_per_k=${recammaster_scene_limit_per_k})"
64
+ # 35% Syn4D dynamic multi-camera mesh-tracked 3D.
65
+ - "87500 @ GeometryRobustnessWrapper_NVS(Syn4D_Multi_NVS(ROOT='${syn4d_root}', caption_dir='${syn4d_caption_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=87500, track_spatial_stride=1, exclude_scenes=['flying_group'], valid_sequences_file='${syn4d_valid_sequences}', use_mp4_rgb=True, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, return_dynamic_masks=True, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}'), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
66
+ # 5% Kubric 49-frame dynamic data.
67
+ - "12500 @ GeometryRobustnessWrapper_NVS(Kubric_NVS(ROOT='${kubric49_root}', caption_cache_dir='${kubric49_caption_cache_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=12500, track_spatial_stride=1, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, return_dynamic_masks=True, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}'), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
68
+ # 5% Kubric 81-frame dynamic data sampled to 49.
69
+ - "12500 @ GeometryRobustnessWrapper_NVS(Kubric_NVS(ROOT='${kubric81_root}', caption_cache_dir='${kubric81_caption_cache_dir}', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=12500, track_spatial_stride=1, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, return_dynamic_masks=True, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}'), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
70
+ # 11.82% DL3DV (real, DA3 depth). Static total is 20%; this keeps the old static relative weight.
71
+ - "29550 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${dl3dv_root}', dataset_type='dl3dv', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=29550, track_spatial_stride=1, min_overlap_ratio=0.15, min_shift_ratio=0.15, max_shift_ratio=0.5, shift_beta_ab=[2.0, 5.0], src_center_in_target_prob=0.5, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, use_da3_depth=True), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
72
+ # 1.38% RE10K (real, DA3 depth).
73
+ - "3450 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${re10k_root}', dataset_type='re10k', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=3450, track_spatial_stride=1, min_overlap_ratio=0.15, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
74
+ # 2.28% MVS-Synth (synthetic, GT depth; scene 0000 held out).
75
+ - "5700 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${mvs_synth_root}', dataset_type='mvs_synth', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=5700, track_spatial_stride=1, min_overlap_ratio=0.15, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True, exclude_sequence_ids=['0000']), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
76
+ # 4.52% TartanAir (static forward-moving scenes).
77
+ - "11300 @ GeometryRobustnessWrapper_NVS(StaticScene_NVS(ROOT='${tartanair_root}', dataset_type='tartanair', num_views=${num_views}, resolution=${resolution}, steps_per_epoch=11300, track_spatial_stride=1, min_overlap_ratio=0.15, min_shift_ratio=0.15, max_shift_ratio=0.5, shift_beta_ab=[2.0, 5.0], src_center_in_target_prob=1.0, return_source_depth_video=True, return_source_rgb_video=True, return_source_cameras=True), sparse_track_prob=${sparse_track_prob}, sparse_track_stride='${sparse_track_stride}', sparse_track_query_stride_mode='${sparse_track_query_stride_mode}', sparse_track_block_jitter_prob=${sparse_track_block_jitter_prob}, sparse_track_block_size=${sparse_track_block_size}, multi_query_prob=${multi_query_prob}, multi_query_choices='${multi_query_choices}', noise_prob=${geometry_noise_prob}, track_offset_std=${track_offset_std}, track_temporal_jitter_std=${track_temporal_jitter_std}, track_xy_noise_ratio=${track_xy_noise_ratio}, track_frame_depth_jitter_ratio=${track_frame_depth_jitter_ratio}, depth_offset_std=${depth_offset_std}, depth_scale_jitter=${depth_scale_jitter}, source_rgb_noise_std=${source_rgb_noise_std}, source_rgb_frame_jitter_std=${source_rgb_frame_jitter_std}, dynamic_mask_aug_prob=${dynamic_mask_aug_prob}, dynamic_mask_morph_kernel=${dynamic_mask_morph_kernel}, dynamic_mask_flip_prob=${dynamic_mask_flip_prob})"
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/torch_profiler/rank0.1780391935713307947.pt.trace.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6046891542cb397d3092b7489775e48a8cbd2749d633b85ee960cc5ee89dab43
3
+ size 4806779109
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/debug-internal.log ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"time":"2026-06-02T09:15:03.360626307Z","level":"INFO","msg":"wandb-core"}
2
+ {"time":"2026-06-02T09:15:03.363384731Z","level":"INFO","msg":"stream: starting","core version":"0.26.0"}
3
+ {"time":"2026-06-02T09:15:03.763480731Z","level":"INFO","msg":"stream: created new stream","id":"iwxrpu9h"}
4
+ {"time":"2026-06-02T09:15:03.76358316Z","level":"INFO","msg":"handler: started"}
5
+ {"time":"2026-06-02T09:15:03.764307429Z","level":"INFO","msg":"stream: started"}
6
+ {"time":"2026-06-02T09:15:03.764319332Z","level":"INFO","msg":"writer: started","stream_id":"iwxrpu9h"}
7
+ {"time":"2026-06-02T09:15:03.764409442Z","level":"INFO","msg":"sender: started"}
8
+ {"time":"2026-06-02T09:15:05.306162953Z","level":"INFO","msg":"filestream: sending request","total_files":1,"console_offset":0,"console_lines":1}
9
+ {"time":"2026-06-02T09:15:05.54706583Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
10
+ {"time":"2026-06-02T09:15:12.213355249Z","level":"INFO","msg":"flowcontrol: backed up, offloading to disk","recordNumber":44}
11
+ {"time":"2026-06-02T09:15:12.218119827Z","level":"INFO","msg":"flowcontrol: unblocked","totalOffloaded":187}
12
+ {"time":"2026-06-02T09:15:20.30658052Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":0,"events_lines":2,"console_offset":1,"console_lines":133,"uploaded_len":2}
13
+ {"time":"2026-06-02T09:15:20.50100822Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
14
+ {"time":"2026-06-02T09:15:35.307091233Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":2,"events_lines":2}
15
+ {"time":"2026-06-02T09:15:35.619078665Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
16
+ {"time":"2026-06-02T09:15:50.306475896Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":4,"events_lines":2,"console_offset":134,"console_lines":4}
17
+ {"time":"2026-06-02T09:15:50.487338735Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
18
+ {"time":"2026-06-02T09:16:05.306770422Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":0,"history_lines":1,"events_offset":6,"events_lines":2,"console_offset":133,"console_lines":1}
19
+ {"time":"2026-06-02T09:16:05.521198844Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
20
+ {"time":"2026-06-02T09:16:20.306562945Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":1,"history_lines":1,"events_offset":8,"events_lines":2,"console_offset":133,"console_lines":1}
21
+ {"time":"2026-06-02T09:16:20.500086028Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
22
+ {"time":"2026-06-02T09:16:35.306929037Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":2,"history_lines":2,"events_offset":10,"events_lines":2,"console_offset":133,"console_lines":1}
23
+ {"time":"2026-06-02T09:16:35.544966706Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
24
+ {"time":"2026-06-02T09:16:50.306544421Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":4,"history_lines":1,"events_offset":12,"events_lines":2,"console_offset":133,"console_lines":1}
25
+ {"time":"2026-06-02T09:16:50.521662086Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
26
+ {"time":"2026-06-02T09:17:05.307011463Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":5,"history_lines":1,"events_offset":14,"events_lines":2,"console_offset":133,"console_lines":1}
27
+ {"time":"2026-06-02T09:17:05.513840193Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
28
+ {"time":"2026-06-02T09:17:20.306873464Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":6,"history_lines":1,"events_offset":16,"events_lines":2,"console_offset":133,"console_lines":1}
29
+ {"time":"2026-06-02T09:17:20.524164031Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
30
+ {"time":"2026-06-02T09:17:35.306466717Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":7,"history_lines":1,"events_offset":18,"events_lines":2,"console_offset":133,"console_lines":1}
31
+ {"time":"2026-06-02T09:17:35.513725414Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
32
+ {"time":"2026-06-02T09:17:50.306691479Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":8,"history_lines":1,"events_offset":20,"events_lines":2,"console_offset":133,"console_lines":1}
33
+ {"time":"2026-06-02T09:17:50.509930633Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
34
+ {"time":"2026-06-02T09:18:05.306781235Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":22,"events_lines":2}
35
+ {"time":"2026-06-02T09:18:05.487008517Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
36
+ {"time":"2026-06-02T09:18:20.306238049Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":24,"events_lines":2}
37
+ {"time":"2026-06-02T09:18:20.594952487Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
38
+ {"time":"2026-06-02T09:18:35.307123907Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":26,"events_lines":2}
39
+ {"time":"2026-06-02T09:18:35.519206589Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
40
+ {"time":"2026-06-02T09:18:50.307778255Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":28,"events_lines":2}
41
+ {"time":"2026-06-02T09:18:50.697208883Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
42
+ {"time":"2026-06-02T09:19:05.306718696Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":30,"events_lines":2}
43
+ {"time":"2026-06-02T09:19:05.753941277Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
44
+ {"time":"2026-06-02T09:19:20.306716421Z","level":"INFO","msg":"filestream: sending request","total_files":1,"events_offset":32,"events_lines":2}
45
+ {"time":"2026-06-02T09:19:20.616359066Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
46
+ {"time":"2026-06-02T09:19:35.307089272Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":34,"events_lines":2,"console_offset":133,"console_lines":1}
47
+ {"time":"2026-06-02T09:19:35.825088137Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
48
+ {"time":"2026-06-02T09:19:50.306659894Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":9,"history_lines":1,"events_offset":36,"events_lines":2,"console_offset":133,"console_lines":1}
49
+ {"time":"2026-06-02T09:19:50.705064787Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
50
+ {"time":"2026-06-02T09:20:05.306909037Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":10,"history_lines":1,"events_offset":38,"events_lines":2,"console_offset":133,"console_lines":1}
51
+ {"time":"2026-06-02T09:20:06.265337713Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
52
+ {"time":"2026-06-02T09:20:20.307661911Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":11,"history_lines":2,"events_offset":40,"events_lines":2,"console_offset":133,"console_lines":1}
53
+ {"time":"2026-06-02T09:20:20.531282276Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
54
+ {"time":"2026-06-02T09:20:35.306992262Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":13,"history_lines":1,"events_offset":42,"events_lines":2,"console_offset":133,"console_lines":1}
55
+ {"time":"2026-06-02T09:20:35.564305086Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
56
+ {"time":"2026-06-02T09:20:50.307023536Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":14,"history_lines":1,"events_offset":44,"events_lines":2,"console_offset":133,"console_lines":1}
57
+ {"time":"2026-06-02T09:20:50.635389586Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
58
+ {"time":"2026-06-02T09:21:05.307044407Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":15,"history_lines":1,"events_offset":46,"events_lines":2,"console_offset":133,"console_lines":1}
59
+ {"time":"2026-06-02T09:21:05.583228743Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
60
+ {"time":"2026-06-02T09:21:20.307285592Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":16,"history_lines":2,"events_offset":48,"events_lines":2,"console_offset":133,"console_lines":1}
61
+ {"time":"2026-06-02T09:21:20.508409608Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
62
+ {"time":"2026-06-02T09:21:35.306917129Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":18,"history_lines":1,"events_offset":50,"events_lines":2,"console_offset":133,"console_lines":1}
63
+ {"time":"2026-06-02T09:21:35.509511474Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
64
+ {"time":"2026-06-02T09:21:50.307082921Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":19,"history_lines":1,"events_offset":52,"events_lines":2,"console_offset":133,"console_lines":1}
65
+ {"time":"2026-06-02T09:21:50.495889909Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
66
+ {"time":"2026-06-02T09:22:05.307152941Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":20,"history_lines":2,"events_offset":54,"events_lines":2,"console_offset":133,"console_lines":1}
67
+ {"time":"2026-06-02T09:22:05.500531686Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
68
+ {"time":"2026-06-02T09:22:20.307303758Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":22,"history_lines":1,"events_offset":56,"events_lines":2,"console_offset":133,"console_lines":1}
69
+ {"time":"2026-06-02T09:22:20.526745051Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
70
+ {"time":"2026-06-02T09:22:35.307493135Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":23,"history_lines":1,"events_offset":58,"events_lines":2,"console_offset":133,"console_lines":1}
71
+ {"time":"2026-06-02T09:22:35.506899762Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
72
+ {"time":"2026-06-02T09:22:50.306876249Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":24,"history_lines":2,"events_offset":60,"events_lines":2,"console_offset":133,"console_lines":1}
73
+ {"time":"2026-06-02T09:22:50.50747968Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
74
+ {"time":"2026-06-02T09:23:05.307030151Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":26,"history_lines":1,"events_offset":62,"events_lines":2,"console_offset":133,"console_lines":1}
75
+ {"time":"2026-06-02T09:23:05.506590001Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
76
+ {"time":"2026-06-02T09:23:20.306698019Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":27,"history_lines":1,"events_offset":64,"events_lines":2,"console_offset":133,"console_lines":1}
77
+ {"time":"2026-06-02T09:23:20.500838005Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
78
+ {"time":"2026-06-02T09:23:35.306631161Z","level":"INFO","msg":"filestream: sending request","total_files":4,"history_offset":28,"history_lines":2,"events_offset":66,"events_lines":2,"console_offset":133,"console_lines":1}
79
+ {"time":"2026-06-02T09:23:35.526000016Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
80
+ {"time":"2026-06-02T09:23:50.30648744Z","level":"INFO","msg":"filestream: sending request","total_files":2,"events_offset":68,"events_lines":2,"console_offset":138,"console_lines":110}
81
+ {"time":"2026-06-02T09:23:50.517760083Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
82
+ {"time":"2026-06-02T09:23:51.771796788Z","level":"INFO","msg":"fileTransfer: Close: file transfer manager closed"}
83
+ {"time":"2026-06-02T09:23:51.772448483Z","level":"INFO","msg":"filestream: sending request","total_files":1,"uploaded_len":3,"complete":true,"exit_code":0}
84
+ {"time":"2026-06-02T09:23:52.050571134Z","level":"INFO","msg":"filestream: request sent","status":"200 OK"}
85
+ {"time":"2026-06-02T09:23:52.056627937Z","level":"INFO","msg":"handler: operation stats","stats":{}}
86
+ {"time":"2026-06-02T09:23:52.063523854Z","level":"INFO","msg":"stream: closing"}
87
+ {"time":"2026-06-02T09:23:52.063589901Z","level":"INFO","msg":"handler: closed"}
88
+ {"time":"2026-06-02T09:23:52.06482974Z","level":"INFO","msg":"sender: closed"}
89
+ {"time":"2026-06-02T09:23:52.064859276Z","level":"INFO","msg":"stream: closed"}
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/debug.log ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2026-06-02 09:15:02,858 INFO MainThread:236392 [wandb_setup.py:_flush():81] Current SDK version is 0.26.0
2
+ 2026-06-02 09:15:02,858 INFO MainThread:236392 [wandb_setup.py:_flush():81] Configure stats pid to 236392
3
+ 2026-06-02 09:15:02,858 INFO MainThread:236392 [wandb_setup.py:_flush():81] Loading settings from environment variables
4
+ 2026-06-02 09:15:02,858 INFO MainThread:236392 [wandb_init.py:setup_run_log_directory():721] Logging user logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/run-20260602_091502-iwxrpu9h/logs/debug.log
5
+ 2026-06-02 09:15:02,859 INFO MainThread:236392 [wandb_init.py:setup_run_log_directory():722] Logging internal logs to /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/run-20260602_091502-iwxrpu9h/logs/debug-internal.log
6
+ 2026-06-02 09:15:02,859 INFO MainThread:236392 [wandb_init.py:init():848] calling init triggers
7
+ 2026-06-02 09:15:02,859 INFO MainThread:236392 [wandb_init.py:init():853] wandb.init called with sweep_config: {}
8
+ config: {'learning_rate': 2e-05, 'weight_decay': 0.01, 'num_epochs': 200, 'save_steps': 500, 'num_workers': 7, 'prefetch_factor': 2, 'dataset_base_path': None, 'dataset_metadata_path': None, 'dataset_repeat': 1, 'dataset_num_workers': 7, 'dataloader_prefetch_factor': 2, 'data_file_keys': 'image,video', 'pointmap_stats_path': None, 'model_paths': None, 'model_id_with_origin_paths': 'Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth', 'model_variant': 'vt2v', 'extra_inputs': None, 'track_xyz_normalization': 'minmax', 'track_xyz_quantile_low': 0.02, 'track_xyz_quantile_high': 0.98, 'predict_track_residual': False, 'residual_target_xyz_source': 'clean', 'residual_zero_center_normalization': False, 'residual_linear_normalization': False, 'residual_log_scale_factor': 3.0, 'stream3r_common_scale_stat': 'mean', 'explicit_motion_loss_weight': 0.0, 'explicit_motion_loss_snr_weight': True, 'explicit_motion_loss_snr_sigmoid': True, 'explicit_motion_loss_snr_sigmoid_a': 1.0, 'explicit_motion_loss_snr_sigmoid_b': -5.0, 'explicit_motion_loss_decode_checkpoint': False, 'explicit_motion_loss_decode_chunk_size': 0, 'normalize_track_with_xyz_cond_stats': False, 'prenormalized_track_video_signed_log': False, 'use_xyz_video_condition': False, 'xyz_video_condition_scale': 1.0, 'xyz_video_condition_aug_prob': 0.0, 'xyz_video_condition_noise_std': 0.0, 'xyz_video_condition_z_noise_std': 0.0, 'xyz_video_condition_frame_scale_std': 0.0, 'xyz_video_condition_dropout_prob': 0.0, 'xyz_video_condition_mode': 'xyz', 'xyz_video_condition_cfg_dropout_prob': 0.0, 'camera_condition_cfg_dropout_prob': 0.0, 'fp8_models': None, 'offload_models': None, 'v2v_concat_mode': 'width', 'v2v_channel_patch_embed_mode': 'add', 'trainable_models': 'dit.domain_embedding,dit.control_patch_embedding', 'find_unused_parameters': False, 'task': 'sft:v2v', 'drop_text_prob': 0.1, 'drop_video_prob': 0.0, 'drop_t2v_prob': None, 'drop_i2v_prob': None, 'input_video_noise_strength': 0.0, 'input_video_noise_random_max': 500, 'output_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout', 'remove_prefix_in_ckpt': 'pipe.dit.', 'save_every_n_epochs': 2, 'max_checkpoints_to_keep': 5, 'allow_existing_checkpoints': True, 'initial_step': 2530, 'max_train_steps': 2560, 'lora_base_model': None, 'lora_target_modules': 'q,k,v,o,ffn.0,ffn.2', 'lora_rank': 32, 'lora_alpha': None, 'lora_checkpoint': None, 'preset_lora_path': None, 'preset_lora_model': None, 'use_gradient_checkpointing': True, 'use_gradient_checkpointing_offload': False, 'gradient_accumulation_steps': 1, 'use_wandb': True, 'wandb_project': 'track4d-360', 'wandb_name': 'ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp', 'wandb_entity': None, 'wandb_tags': ['ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug'], 'wandb_notes': None, 'wandb_log_gradients': False, 'wandb_log_parameters': False, 'wandb_log_freq': 100, 'wandb_watch_model': False, 'debug_latents': False, 'debug_latents_every': 500, 'debug_latents_fps': 16, 'debug_latents_quality': 5, 'debug_latents_max_frames': None, 'debug_latents_outdir': 'debug_latents', 'debug_latents_vae_tiled': True, 'debug_latents_vae_tile_size': (30, 52), 'debug_latents_vae_tile_stride': (15, 26), 'debug_missing_grads': False, 'torch_distributed_debug': None, 'use_query_frame_impulse_condition': True, 'height': 288, 'width': 512, 'max_pixels': 1048576, 'num_frames': 49, 'tokenizer_path': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl', 'dit_finetune_modules': 'self_attn', 'track_injection_mode': 'per_block', 'track_injection_block_mode': 'concat_project', 'train_sigma_shift': 5.0, 'dummy': False, 'steps_per_epoch': 25000, 'initialize_model_on_cpu': False, 'resume_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors', 'gs_head_init_checkpoint': None, 'joint_track_prediction': False, 'joint_track_depth_prediction': False, 'joint_depth_prediction': False, 'target_depth_log_cap': 14.0, 'target_depth_norm_type': 'log_depth', 'target_depth_max_disparity': 5.0, 'vxf_init_checkpoint': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors', 'channel_concat_stage': None, 'use_residual_tracks': True, 'freeze_lora_after_load': False, 'skip_v2v_required_trainables': False, 'gs_only_training': False, 'freeze_track_block_injector': False, 'freeze_track4d_condition_modules': False, 'gs_head_style': 'flashworld', 'use_gs_output': False, 'gs_zero_features': False, 'gs_feat_dim': 1024, 'gs_loss_weight': 1.0, 'gs_mse_loss_weight': 1.0, 'gs_lpips_loss_weight': 0.5, 'gs_depth_loss_weight': 0.05, 'gs_abs_depth_loss_weight': 0.0, 'gs_opacity_loss_weight': 0.1, 'gs_opacity_prune_keep_frac': 1.0, 'gs_rgb_loss_weight': None, 'gs_sh_degree': 2, 'gs_num_points_per_pixel': 2, 'no_gs_latent_detach': False, 'use_gs_feedback_training': False, 'gs_scale_cap': None, 'gs_render_patch_size': None, 'gs_legacy_union_render': False, 'gs_lyra_depth_param': 'exp', 'gs_lyra_log_depth_shift': 0.0, 'gs_lyra_dnear': 0.1, 'gs_lyra_dfar': 10.0, 'gs_replay_cache_size': 0, 'gs_dynamic_mode': 'static', 'gs_track_supervision_mode': 'pointmap', 'gs_pos_loss_weight': 1.0, 'gs_dyn_depth_loss_weight': 0.5, 'gs_opa_tv_weight': 0.01, 'gs_xyz_tv_weight': 0.01, 'gs_predict_rotation': False, 'gs_disable_delta_opacity': False, 'gs_direct_pos_loss_weight': 0.0, 'gs_track_canonical_loss_weight': 0.0, 'gs_track_delta_loss_weight': 0.0, 'gs_track_final_pos_loss_weight': 0.0, 'gs_dynamic_max_t_video': 64, 'gs_assoc_track_2d_loss_weight': 0.0, 'gs_track_accel_loss_weight': 0.0, 'gs_track_loss_quantile': 0.98, 'gs_track_canonical_detach': False, 'gs_track_joint_render': False, 'cond_drop_track_prob': 0.1, 'cond_drop_dense_prob': 0.0, 'cond_drop_geo_prob': 0.1, 'use_track_null_token_fallback': False, 'use_dense_null_token_fallback': False, 'use_plucker_camera_control': True, 'enable_v2v_plucker_camera_control': True, 'train_plucker_camera_control': False, 'use_plucker_raymap_tokens': True, 'plucker_use_ray_origin': False, 'use_perlayer_camera_control': True, 'camera_injection_mode': 'projector', 'camera_plucker_encoder_mode': 'patchify', 'warped_condition_mode': 'off', 'warped_appearance_fusion': 'concat_proj', 'warped_geom_only_dense': True, 'multi_frame_warped_appearance': False, 'multi_frame_motion_threshold': 0.01, 'fixed_step_frac': None, 'fixed_num_inference_steps': 50, 'fixed_sigma': None, 'skip_diffusion_loss': False, 'use_dense_branch': True, 'dense_in_channels': 5, 'dense_proj_dim': 32, 'dense_num_residual_blocks': 2, 'dense_alpha_track': 1.0, 'use_depth_null_token': False, 'dense_conditioning_noise_max_level': 0.15, 'source_mask_boundary_depth_dropout_prob': 0.5, 'source_mask_boundary_kernel_size': 5, 'dense_warp_mode': 'hybrid_softsplat', 'dense_warp_motion_threshold': 0.005, 'dense_hybrid_disable_dynamic_masks': False, 'use_vista4d_latent_streams': False, 'use_gen3c_warp_latent_channels': False, 'disable_track_modules': False, 'dense_only_condition_modules': False, 'torch_profile_active_steps': 5, 'torch_profile_wait_steps': 3, 'torch_profile_warmup_steps': 1, 'torch_profile_repeat': 1, 'torch_profile_dir': None, 'torch_profile_record_shapes': False, 'torch_profile_memory': False, 'torch_profile_with_stack': False, 'track_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth.yaml', 'track4d360_data_config': '/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml', 'syn4d_root': '/scratch/shared/beegfs/zeren/Syn4D', 'scene': None, 'sequence_idx': None, 'track_spatial_stride': 1, '_wandb': {}}
9
+ 2026-06-02 09:15:02,859 INFO MainThread:236392 [wandb_init.py:init():896] starting backend
10
+ 2026-06-02 09:15:03,349 INFO MainThread:236392 [wandb_init.py:init():911] sending inform_init request
11
+ 2026-06-02 09:15:03,357 INFO MainThread:236392 [wandb_init.py:init():919] backend started and connected
12
+ 2026-06-02 09:15:03,360 INFO MainThread:236392 [wandb_init.py:init():989] updated telemetry
13
+ 2026-06-02 09:15:03,426 INFO MainThread:236392 [wandb_init.py:init():1013] communicating run to backend with 90.0 second timeout
14
+ 2026-06-02 09:15:04,130 INFO MainThread:236392 [wandb_init.py:init():1058] starting run threads in backend
15
+ 2026-06-02 09:15:05,292 INFO MainThread:236392 [wandb_run.py:_console_start():2542] atexit reg
16
+ 2026-06-02 09:15:05,292 INFO MainThread:236392 [wandb_run.py:_redirect():2391] redirect: wrap_raw
17
+ 2026-06-02 09:15:05,293 INFO MainThread:236392 [wandb_run.py:_redirect():2460] Wrapping output streams.
18
+ 2026-06-02 09:15:05,293 INFO MainThread:236392 [wandb_run.py:_redirect():2483] Redirects installed.
19
+ 2026-06-02 09:15:05,299 INFO MainThread:236392 [wandb_init.py:init():1098] run started, returning control to user process
20
+ 2026-06-02 09:23:51,064 INFO MainThread:236392 [wandb_run.py:_finish():2309] finishing run nirvanalan/track4d-360/iwxrpu9h
21
+ 2026-06-02 09:23:51,067 INFO MainThread:236392 [wandb_run.py:_atexit_cleanup():2508] got exitcode: 0
22
+ 2026-06-02 09:23:51,068 INFO MainThread:236392 [wandb_run.py:_restore():2490] restore
23
+ 2026-06-02 09:23:51,069 INFO MainThread:236392 [wandb_run.py:_restore():2496] restore done
24
+ 2026-06-02 09:23:52,062 INFO MainThread:236392 [wandb_run.py:_footer_sync_info():3886] logging synced files
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/run-20260602_091502-iwxrpu9h/files/config.yaml ADDED
@@ -0,0 +1,700 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ _wandb:
2
+ value:
3
+ cli_version: 0.26.0
4
+ e:
5
+ f0ztrk8t2qi0av8h7ptk8q57xoxkau1o:
6
+ args:
7
+ - --model_id_with_origin_paths
8
+ - Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth
9
+ - --tokenizer_path
10
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl
11
+ - --task
12
+ - sft:v2v
13
+ - --model_variant
14
+ - vt2v
15
+ - --v2v_concat_mode
16
+ - width
17
+ - --height
18
+ - "288"
19
+ - --width
20
+ - "512"
21
+ - --num_frames
22
+ - "49"
23
+ - --track4d360_data_config
24
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml
25
+ - --steps_per_epoch
26
+ - "25000"
27
+ - --num_epochs
28
+ - "200"
29
+ - --learning_rate
30
+ - "2e-5"
31
+ - --gradient_accumulation_steps
32
+ - "1"
33
+ - --save_steps
34
+ - "500"
35
+ - --remove_prefix_in_ckpt
36
+ - pipe.dit.
37
+ - --trainable_models
38
+ - dit.domain_embedding,dit.control_patch_embedding
39
+ - --dit_finetune_modules
40
+ - self_attn
41
+ - --track_injection_mode
42
+ - per_block
43
+ - --track_injection_block_mode
44
+ - concat_project
45
+ - --track_config
46
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth.yaml
47
+ - --train_sigma_shift
48
+ - "5.0"
49
+ - --use_query_frame_impulse_condition
50
+ - --use_plucker_camera_control
51
+ - --enable_v2v_plucker_camera_control
52
+ - --use_plucker_raymap_tokens
53
+ - --use_perlayer_camera_control
54
+ - --camera_injection_mode
55
+ - projector
56
+ - --camera_plucker_encoder_mode
57
+ - patchify
58
+ - --use_dense_branch
59
+ - --dense_in_channels
60
+ - "5"
61
+ - --dense_proj_dim
62
+ - "32"
63
+ - --dense_num_residual_blocks
64
+ - "2"
65
+ - --dense_alpha_track
66
+ - "1.0"
67
+ - --dense_conditioning_noise_max_level
68
+ - "0.15"
69
+ - --source_mask_boundary_depth_dropout_prob
70
+ - "0.5"
71
+ - --source_mask_boundary_kernel_size
72
+ - "5"
73
+ - --dense_warp_mode
74
+ - hybrid_softsplat
75
+ - --dense_warp_motion_threshold
76
+ - "0.005"
77
+ - --cond_drop_geo_prob
78
+ - "0.1"
79
+ - --cond_drop_track_prob
80
+ - "0.1"
81
+ - --cond_drop_dense_prob
82
+ - "0.0"
83
+ - --initial_step
84
+ - "2530"
85
+ - --output_path
86
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout
87
+ - --dataset_num_workers
88
+ - "7"
89
+ - --dataloader_prefetch_factor
90
+ - "2"
91
+ - --drop_text_prob
92
+ - "0.1"
93
+ - --debug_latents_every
94
+ - "500"
95
+ - --allow_existing_checkpoints
96
+ - --wandb_project
97
+ - track4d-360
98
+ - --wandb_name
99
+ - ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp
100
+ - --wandb_tags
101
+ - ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug
102
+ - --use_wandb
103
+ - --vxf_init_checkpoint
104
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors
105
+ - --use_gradient_checkpointing
106
+ - --resume_checkpoint
107
+ - /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors
108
+ - --max_train_steps
109
+ - "2560"
110
+ - --torch_profile_active_steps
111
+ - "5"
112
+ - --torch_profile_wait_steps
113
+ - "3"
114
+ - --torch_profile_warmup_steps
115
+ - "1"
116
+ - --torch_profile_repeat
117
+ - "1"
118
+ - --no_torch_profile_record_shapes
119
+ - --no_torch_profile_memory
120
+ codePath: scripts/wan2.1_fun/train_track4d360.py
121
+ codePathLocal: scripts/wan2.1_fun/train_track4d360.py
122
+ cpu_count: 288
123
+ cpu_count_logical: 288
124
+ cudaVersion: "12.7"
125
+ disk:
126
+ /:
127
+ total: "255795724288"
128
+ used: "549191680"
129
+ email: yushi.lan@outlook.com
130
+ executable: /projects/public/u6ew/yslan_public/yslan_project/miniforge3/envs/th29py312/bin/python3.12
131
+ git:
132
+ commit: e47380b0c0f4cbb87efcbad540b96a404c2ccc88
133
+ remote: git@github.com:NIRVANALAN/track4d-360.git
134
+ gpu: NVIDIA GH200 120GB
135
+ gpu_count: 4
136
+ gpu_nvidia:
137
+ - architecture: Hopper
138
+ cudaCores: 16896
139
+ memoryTotal: "102625181696"
140
+ name: NVIDIA GH200 120GB
141
+ uuid: GPU-d655252a-055e-6123-1fcc-96c384e94cb6
142
+ - architecture: Hopper
143
+ cudaCores: 16896
144
+ memoryTotal: "102625181696"
145
+ name: NVIDIA GH200 120GB
146
+ uuid: GPU-f32c462b-35e6-cd39-3ab1-ef2d12935745
147
+ - architecture: Hopper
148
+ cudaCores: 16896
149
+ memoryTotal: "102625181696"
150
+ name: NVIDIA GH200 120GB
151
+ uuid: GPU-c6448e24-bbf3-db8b-e6c7-7becaf680d11
152
+ - architecture: Hopper
153
+ cudaCores: 16896
154
+ memoryTotal: "102625181696"
155
+ name: NVIDIA GH200 120GB
156
+ uuid: GPU-2c148b70-ecf2-fb15-969b-cbb11a8b9bc4
157
+ host: nid011121
158
+ memory:
159
+ total: "919613407232"
160
+ os: Linux-6.4.0-150600.23.73_15.0.14-cray_shasta_c_64k-aarch64-with-glibc2.38
161
+ program: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/scripts/wan2.1_fun/train_track4d360.py
162
+ python: CPython 3.12.13
163
+ root: /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout
164
+ slurm:
165
+ cluster_name: gracehopper
166
+ conf: /var/spool/slurmd/conf-cache/slurm.conf
167
+ cpu_bind: quiet,mask_cpu:0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
168
+ cpu_bind_list: 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF
169
+ cpu_bind_type: 'mask_cpu:'
170
+ cpu_bind_verbose: quiet
171
+ cpus_on_node: "288"
172
+ distribution: cyclic
173
+ gpus: "32"
174
+ gpus_on_node: "4"
175
+ gpus_per_node: "4"
176
+ gtids: "0"
177
+ job_account: brics.u6ew
178
+ job_cpus_per_node: 288(x8)
179
+ job_end_time: "1780398532"
180
+ job_gid: "1483804847"
181
+ job_gpus: 0,1,2,3
182
+ job_id: "4984787"
183
+ job_name: 14b_pair1x49_profile_patchifylinear_fsdp2
184
+ job_nodelist: nid[011121-011122,011125,011127-011129,011165,011167]
185
+ job_num_nodes: "8"
186
+ job_partition: workq
187
+ job_qos: normal
188
+ job_start_time: "1780391332"
189
+ job_uid: "1483804847"
190
+ job_user: yushi.u6ew
191
+ jobid: "4984787"
192
+ launch_node_ipaddr: 10.242.4.57
193
+ localid: "0"
194
+ mpi_type: pmi2
195
+ nnodes: "8"
196
+ nodeid: "0"
197
+ nodelist: nid[011121-011122,011125,011127-011129,011165,011167]
198
+ nprocs: "8"
199
+ ntasks: "8"
200
+ ntasks_per_node: "1"
201
+ oom_kill_step: "0"
202
+ prio_process: "0"
203
+ procid: "0"
204
+ srun_comm_host: 10.242.4.57
205
+ srun_comm_port: "45551"
206
+ step_gpus: 0,1,2,3
207
+ step_id: "0"
208
+ step_launcher_port: "45551"
209
+ step_nodelist: nid[011121-011122,011125,011127-011129,011165,011167]
210
+ step_num_nodes: "8"
211
+ step_num_tasks: "8"
212
+ step_resv_ports: 27631-27632
213
+ step_tasks_per_node: 1(x8)
214
+ stepid: "0"
215
+ submit_dir: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360
216
+ submit_host: login40
217
+ task_pid: "236308"
218
+ tasks_per_node: 1(x8)
219
+ topology_addr: global.group12.nid011121
220
+ topology_addr_pattern: switch.switch.node
221
+ umask: "0022"
222
+ startedAt: "2026-06-02T09:15:02.848946Z"
223
+ writerId: f0ztrk8t2qi0av8h7ptk8q57xoxkau1o
224
+ m: []
225
+ python_version: 3.12.13
226
+ t:
227
+ "1":
228
+ - 1
229
+ - 11
230
+ - 41
231
+ - 49
232
+ - 71
233
+ - 83
234
+ - 98
235
+ - 106
236
+ "2":
237
+ - 1
238
+ - 11
239
+ - 41
240
+ - 49
241
+ - 51
242
+ - 71
243
+ - 83
244
+ - 98
245
+ - 106
246
+ "3":
247
+ - 2
248
+ - 13
249
+ - 15
250
+ - 16
251
+ - 41
252
+ - 61
253
+ "4": 3.12.13
254
+ "5": 0.26.0
255
+ "6": 4.46.2
256
+ "12": 0.26.0
257
+ "13": linux-aarch64
258
+ allow_existing_checkpoints:
259
+ value: true
260
+ camera_condition_cfg_dropout_prob:
261
+ value: 0
262
+ camera_injection_mode:
263
+ value: projector
264
+ camera_plucker_encoder_mode:
265
+ value: patchify
266
+ channel_concat_stage:
267
+ value: null
268
+ cond_drop_dense_prob:
269
+ value: 0
270
+ cond_drop_geo_prob:
271
+ value: 0.1
272
+ cond_drop_track_prob:
273
+ value: 0.1
274
+ data_file_keys:
275
+ value: image,video
276
+ dataloader_prefetch_factor:
277
+ value: 2
278
+ dataset_base_path:
279
+ value: null
280
+ dataset_metadata_path:
281
+ value: null
282
+ dataset_num_workers:
283
+ value: 7
284
+ dataset_repeat:
285
+ value: 1
286
+ debug_latents:
287
+ value: false
288
+ debug_latents_every:
289
+ value: 500
290
+ debug_latents_fps:
291
+ value: 16
292
+ debug_latents_max_frames:
293
+ value: null
294
+ debug_latents_outdir:
295
+ value: debug_latents
296
+ debug_latents_quality:
297
+ value: 5
298
+ debug_latents_vae_tile_size:
299
+ value:
300
+ - 30
301
+ - 52
302
+ debug_latents_vae_tile_stride:
303
+ value:
304
+ - 15
305
+ - 26
306
+ debug_latents_vae_tiled:
307
+ value: true
308
+ debug_missing_grads:
309
+ value: false
310
+ dense_alpha_track:
311
+ value: 1
312
+ dense_conditioning_noise_max_level:
313
+ value: 0.15
314
+ dense_hybrid_disable_dynamic_masks:
315
+ value: false
316
+ dense_in_channels:
317
+ value: 5
318
+ dense_num_residual_blocks:
319
+ value: 2
320
+ dense_only_condition_modules:
321
+ value: false
322
+ dense_proj_dim:
323
+ value: 32
324
+ dense_warp_mode:
325
+ value: hybrid_softsplat
326
+ dense_warp_motion_threshold:
327
+ value: 0.005
328
+ disable_track_modules:
329
+ value: false
330
+ dit_finetune_modules:
331
+ value: self_attn
332
+ drop_i2v_prob:
333
+ value: null
334
+ drop_t2v_prob:
335
+ value: null
336
+ drop_text_prob:
337
+ value: 0.1
338
+ drop_video_prob:
339
+ value: 0
340
+ dummy:
341
+ value: false
342
+ enable_v2v_plucker_camera_control:
343
+ value: true
344
+ explicit_motion_loss_decode_checkpoint:
345
+ value: false
346
+ explicit_motion_loss_decode_chunk_size:
347
+ value: 0
348
+ explicit_motion_loss_snr_sigmoid:
349
+ value: true
350
+ explicit_motion_loss_snr_sigmoid_a:
351
+ value: 1
352
+ explicit_motion_loss_snr_sigmoid_b:
353
+ value: -5
354
+ explicit_motion_loss_snr_weight:
355
+ value: true
356
+ explicit_motion_loss_weight:
357
+ value: 0
358
+ extra_inputs:
359
+ value: null
360
+ find_unused_parameters:
361
+ value: false
362
+ fixed_num_inference_steps:
363
+ value: 50
364
+ fixed_sigma:
365
+ value: null
366
+ fixed_step_frac:
367
+ value: null
368
+ fp8_models:
369
+ value: null
370
+ freeze_lora_after_load:
371
+ value: false
372
+ freeze_track_block_injector:
373
+ value: false
374
+ freeze_track4d_condition_modules:
375
+ value: false
376
+ gradient_accumulation_steps:
377
+ value: 1
378
+ gs_abs_depth_loss_weight:
379
+ value: 0
380
+ gs_assoc_track_2d_loss_weight:
381
+ value: 0
382
+ gs_depth_loss_weight:
383
+ value: 0.05
384
+ gs_direct_pos_loss_weight:
385
+ value: 0
386
+ gs_disable_delta_opacity:
387
+ value: false
388
+ gs_dyn_depth_loss_weight:
389
+ value: 0.5
390
+ gs_dynamic_max_t_video:
391
+ value: 64
392
+ gs_dynamic_mode:
393
+ value: static
394
+ gs_feat_dim:
395
+ value: 1024
396
+ gs_head_init_checkpoint:
397
+ value: null
398
+ gs_head_style:
399
+ value: flashworld
400
+ gs_legacy_union_render:
401
+ value: false
402
+ gs_loss_weight:
403
+ value: 1
404
+ gs_lpips_loss_weight:
405
+ value: 0.5
406
+ gs_lyra_depth_param:
407
+ value: exp
408
+ gs_lyra_dfar:
409
+ value: 10
410
+ gs_lyra_dnear:
411
+ value: 0.1
412
+ gs_lyra_log_depth_shift:
413
+ value: 0
414
+ gs_mse_loss_weight:
415
+ value: 1
416
+ gs_num_points_per_pixel:
417
+ value: 2
418
+ gs_only_training:
419
+ value: false
420
+ gs_opa_tv_weight:
421
+ value: 0.01
422
+ gs_opacity_loss_weight:
423
+ value: 0.1
424
+ gs_opacity_prune_keep_frac:
425
+ value: 1
426
+ gs_pos_loss_weight:
427
+ value: 1
428
+ gs_predict_rotation:
429
+ value: false
430
+ gs_render_patch_size:
431
+ value: null
432
+ gs_replay_cache_size:
433
+ value: 0
434
+ gs_rgb_loss_weight:
435
+ value: null
436
+ gs_scale_cap:
437
+ value: null
438
+ gs_sh_degree:
439
+ value: 2
440
+ gs_track_accel_loss_weight:
441
+ value: 0
442
+ gs_track_canonical_detach:
443
+ value: false
444
+ gs_track_canonical_loss_weight:
445
+ value: 0
446
+ gs_track_delta_loss_weight:
447
+ value: 0
448
+ gs_track_final_pos_loss_weight:
449
+ value: 0
450
+ gs_track_joint_render:
451
+ value: false
452
+ gs_track_loss_quantile:
453
+ value: 0.98
454
+ gs_track_supervision_mode:
455
+ value: pointmap
456
+ gs_xyz_tv_weight:
457
+ value: 0.01
458
+ gs_zero_features:
459
+ value: false
460
+ height:
461
+ value: 288
462
+ initial_step:
463
+ value: 2530
464
+ initialize_model_on_cpu:
465
+ value: false
466
+ input_video_noise_random_max:
467
+ value: 500
468
+ input_video_noise_strength:
469
+ value: 0
470
+ joint_depth_prediction:
471
+ value: false
472
+ joint_track_depth_prediction:
473
+ value: false
474
+ joint_track_prediction:
475
+ value: false
476
+ learning_rate:
477
+ value: 2e-05
478
+ lora_alpha:
479
+ value: null
480
+ lora_base_model:
481
+ value: null
482
+ lora_checkpoint:
483
+ value: null
484
+ lora_rank:
485
+ value: 32
486
+ lora_target_modules:
487
+ value: q,k,v,o,ffn.0,ffn.2
488
+ max_checkpoints_to_keep:
489
+ value: 5
490
+ max_pixels:
491
+ value: 1048576
492
+ max_train_steps:
493
+ value: 2560
494
+ model_id_with_origin_paths:
495
+ value: Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth
496
+ model_paths:
497
+ value: null
498
+ model_variant:
499
+ value: vt2v
500
+ multi_frame_motion_threshold:
501
+ value: 0.01
502
+ multi_frame_warped_appearance:
503
+ value: false
504
+ no_gs_latent_detach:
505
+ value: false
506
+ normalize_track_with_xyz_cond_stats:
507
+ value: false
508
+ num_epochs:
509
+ value: 200
510
+ num_frames:
511
+ value: 49
512
+ num_workers:
513
+ value: 7
514
+ offload_models:
515
+ value: null
516
+ output_path:
517
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout
518
+ plucker_use_ray_origin:
519
+ value: false
520
+ pointmap_stats_path:
521
+ value: null
522
+ predict_track_residual:
523
+ value: false
524
+ prefetch_factor:
525
+ value: 2
526
+ prenormalized_track_video_signed_log:
527
+ value: false
528
+ preset_lora_model:
529
+ value: null
530
+ preset_lora_path:
531
+ value: null
532
+ remove_prefix_in_ckpt:
533
+ value: pipe.dit.
534
+ residual_linear_normalization:
535
+ value: false
536
+ residual_log_scale_factor:
537
+ value: 3
538
+ residual_target_xyz_source:
539
+ value: clean
540
+ residual_zero_center_normalization:
541
+ value: false
542
+ resume_checkpoint:
543
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors
544
+ save_every_n_epochs:
545
+ value: 2
546
+ save_steps:
547
+ value: 500
548
+ scene:
549
+ value: null
550
+ sequence_idx:
551
+ value: null
552
+ skip_diffusion_loss:
553
+ value: false
554
+ skip_v2v_required_trainables:
555
+ value: false
556
+ source_mask_boundary_depth_dropout_prob:
557
+ value: 0.5
558
+ source_mask_boundary_kernel_size:
559
+ value: 5
560
+ steps_per_epoch:
561
+ value: 25000
562
+ stream3r_common_scale_stat:
563
+ value: mean
564
+ syn4d_root:
565
+ value: /scratch/shared/beegfs/zeren/Syn4D
566
+ target_depth_log_cap:
567
+ value: 14
568
+ target_depth_max_disparity:
569
+ value: 5
570
+ target_depth_norm_type:
571
+ value: log_depth
572
+ task:
573
+ value: sft:v2v
574
+ tokenizer_path:
575
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl
576
+ torch_distributed_debug:
577
+ value: null
578
+ torch_profile_active_steps:
579
+ value: 5
580
+ torch_profile_dir:
581
+ value: null
582
+ torch_profile_memory:
583
+ value: false
584
+ torch_profile_record_shapes:
585
+ value: false
586
+ torch_profile_repeat:
587
+ value: 1
588
+ torch_profile_wait_steps:
589
+ value: 3
590
+ torch_profile_warmup_steps:
591
+ value: 1
592
+ torch_profile_with_stack:
593
+ value: false
594
+ track_config:
595
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth.yaml
596
+ track_injection_block_mode:
597
+ value: concat_project
598
+ track_injection_mode:
599
+ value: per_block
600
+ track_spatial_stride:
601
+ value: 1
602
+ track_xyz_normalization:
603
+ value: minmax
604
+ track_xyz_quantile_high:
605
+ value: 0.98
606
+ track_xyz_quantile_low:
607
+ value: 0.02
608
+ track4d360_data_config:
609
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml
610
+ train_plucker_camera_control:
611
+ value: false
612
+ train_sigma_shift:
613
+ value: 5
614
+ trainable_models:
615
+ value: dit.domain_embedding,dit.control_patch_embedding
616
+ use_dense_branch:
617
+ value: true
618
+ use_dense_null_token_fallback:
619
+ value: false
620
+ use_depth_null_token:
621
+ value: false
622
+ use_gen3c_warp_latent_channels:
623
+ value: false
624
+ use_gradient_checkpointing:
625
+ value: true
626
+ use_gradient_checkpointing_offload:
627
+ value: false
628
+ use_gs_feedback_training:
629
+ value: false
630
+ use_gs_output:
631
+ value: false
632
+ use_perlayer_camera_control:
633
+ value: true
634
+ use_plucker_camera_control:
635
+ value: true
636
+ use_plucker_raymap_tokens:
637
+ value: true
638
+ use_query_frame_impulse_condition:
639
+ value: true
640
+ use_residual_tracks:
641
+ value: true
642
+ use_track_null_token_fallback:
643
+ value: false
644
+ use_vista4d_latent_streams:
645
+ value: false
646
+ use_wandb:
647
+ value: true
648
+ use_xyz_video_condition:
649
+ value: false
650
+ v2v_channel_patch_embed_mode:
651
+ value: add
652
+ v2v_concat_mode:
653
+ value: width
654
+ vxf_init_checkpoint:
655
+ value: /lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors
656
+ wandb_entity:
657
+ value: null
658
+ wandb_log_freq:
659
+ value: 100
660
+ wandb_log_gradients:
661
+ value: false
662
+ wandb_log_parameters:
663
+ value: false
664
+ wandb_name:
665
+ value: ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp
666
+ wandb_notes:
667
+ value: null
668
+ wandb_project:
669
+ value: track4d-360
670
+ wandb_tags:
671
+ value:
672
+ - ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug
673
+ wandb_watch_model:
674
+ value: false
675
+ warped_appearance_fusion:
676
+ value: concat_proj
677
+ warped_condition_mode:
678
+ value: "off"
679
+ warped_geom_only_dense:
680
+ value: true
681
+ weight_decay:
682
+ value: 0.01
683
+ width:
684
+ value: 512
685
+ xyz_video_condition_aug_prob:
686
+ value: 0
687
+ xyz_video_condition_cfg_dropout_prob:
688
+ value: 0
689
+ xyz_video_condition_dropout_prob:
690
+ value: 0
691
+ xyz_video_condition_frame_scale_std:
692
+ value: 0
693
+ xyz_video_condition_mode:
694
+ value: xyz
695
+ xyz_video_condition_noise_std:
696
+ value: 0
697
+ xyz_video_condition_scale:
698
+ value: 1
699
+ xyz_video_condition_z_noise_std:
700
+ value: 0
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/run-20260602_091502-iwxrpu9h/files/output.log ADDED
@@ -0,0 +1,248 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ✓ WandB initialized: https://wandb.ai/nirvanalan/track4d-360/runs/iwxrpu9h
2
+ /projects/public/u6ew/yslan_public/yslan_project/miniforge3/envs/th29py312/lib/python3.12/site-packages/accelerate/utils/fsdp_utils.py:742: UserWarning: FSDP upcast of low precision parameters to fp32 (since mixed_precision != 'no') may affect the precision of model checkpoints. This effects 597 parameters: ['pipe.dit.domain_embedding', 'pipe.dit.blocks.0.self_attn.q.weight', 'pipe.dit.blocks.0.self_attn.q.bias', 'pipe.dit.blocks.0.self_attn.k.weight', 'pipe.dit.blocks.0.self_attn.k.bias', 'pipe.dit.blocks.0.self_attn.v.weight', 'pipe.dit.blocks.0.self_attn.v.bias', 'pipe.dit.blocks.0.self_attn.o.weight', 'pipe.dit.blocks.0.self_attn.o.bias', 'pipe.dit.blocks.0.self_attn.norm_q.weight', 'pipe.dit.blocks.0.self_attn.norm_k.weight', 'pipe.dit.blocks.0.camera_projector.weight', 'pipe.dit.blocks.0.camera_projector.bias', 'pipe.dit.blocks.1.self_attn.q.weight', 'pipe.dit.blocks.1.self_attn.q.bias', 'pipe.dit.blocks.1.self_attn.k.weight', 'pipe.dit.blocks.1.self_attn.k.bias', 'pipe.dit.blocks.1.self_attn.v.weight', 'pipe.dit.blocks.1.self_attn.v.bias', 'pipe.dit.blocks.1.self_attn.o.weight', 'pipe.dit.blocks.1.self_attn.o.bias', 'pipe.dit.blocks.1.self_attn.norm_q.weight', 'pipe.dit.blocks.1.self_attn.norm_k.weight', 'pipe.dit.blocks.1.camera_projector.weight', 'pipe.dit.blocks.1.camera_projector.bias', 'pipe.dit.blocks.2.self_attn.q.weight', 'pipe.dit.blocks.2.self_attn.q.bias', 'pipe.dit.blocks.2.self_attn.k.weight', 'pipe.dit.blocks.2.self_attn.k.bias', 'pipe.dit.blocks.2.self_attn.v.weight', 'pipe.dit.blocks.2.self_attn.v.bias', 'pipe.dit.blocks.2.self_attn.o.weight', 'pipe.dit.blocks.2.self_attn.o.bias', 'pipe.dit.blocks.2.self_attn.norm_q.weight', 'pipe.dit.blocks.2.self_attn.norm_k.weight', 'pipe.dit.blocks.2.camera_projector.weight', 'pipe.dit.blocks.2.camera_projector.bias', 'pipe.dit.blocks.3.self_attn.q.weight', 'pipe.dit.blocks.3.self_attn.q.bias', 'pipe.dit.blocks.3.self_attn.k.weight', 'pipe.dit.blocks.3.self_attn.k.bias', 'pipe.dit.blocks.3.self_attn.v.weight', 'pipe.dit.blocks.3.self_attn.v.bias', 'pipe.dit.blocks.3.self_attn.o.weight', 'pipe.dit.blocks.3.self_attn.o.bias', 'pipe.dit.blocks.3.self_attn.norm_q.weight', 'pipe.dit.blocks.3.self_attn.norm_k.weight', 'pipe.dit.blocks.3.camera_projector.weight', 'pipe.dit.blocks.3.camera_projector.bias', 'pipe.dit.blocks.4.self_attn.q.weight', 'pipe.dit.blocks.4.self_attn.q.bias', 'pipe.dit.blocks.4.self_attn.k.weight', 'pipe.dit.blocks.4.self_attn.k.bias', 'pipe.dit.blocks.4.self_attn.v.weight', 'pipe.dit.blocks.4.self_attn.v.bias', 'pipe.dit.blocks.4.self_attn.o.weight', 'pipe.dit.blocks.4.self_attn.o.bias', 'pipe.dit.blocks.4.self_attn.norm_q.weight', 'pipe.dit.blocks.4.self_attn.norm_k.weight', 'pipe.dit.blocks.4.camera_projector.weight', 'pipe.dit.blocks.4.camera_projector.bias', 'pipe.dit.blocks.5.self_attn.q.weight', 'pipe.dit.blocks.5.self_attn.q.bias', 'pipe.dit.blocks.5.self_attn.k.weight', 'pipe.dit.blocks.5.self_attn.k.bias', 'pipe.dit.blocks.5.self_attn.v.weight', 'pipe.dit.blocks.5.self_attn.v.bias', 'pipe.dit.blocks.5.self_attn.o.weight', 'pipe.dit.blocks.5.self_attn.o.bias', 'pipe.dit.blocks.5.self_attn.norm_q.weight', 'pipe.dit.blocks.5.self_attn.norm_k.weight', 'pipe.dit.blocks.5.camera_projector.weight', 'pipe.dit.blocks.5.camera_projector.bias', 'pipe.dit.blocks.6.self_attn.q.weight', 'pipe.dit.blocks.6.self_attn.q.bias', 'pipe.dit.blocks.6.self_attn.k.weight', 'pipe.dit.blocks.6.self_attn.k.bias', 'pipe.dit.blocks.6.self_attn.v.weight', 'pipe.dit.blocks.6.self_attn.v.bias', 'pipe.dit.blocks.6.self_attn.o.weight', 'pipe.dit.blocks.6.self_attn.o.bias', 'pipe.dit.blocks.6.self_attn.norm_q.weight', 'pipe.dit.blocks.6.self_attn.norm_k.weight', 'pipe.dit.blocks.6.camera_projector.weight', 'pipe.dit.blocks.6.camera_projector.bias', 'pipe.dit.blocks.7.self_attn.q.weight', 'pipe.dit.blocks.7.self_attn.q.bias', 'pipe.dit.blocks.7.self_attn.k.weight', 'pipe.dit.blocks.7.self_attn.k.bias', 'pipe.dit.blocks.7.self_attn.v.weight', 'pipe.dit.blocks.7.self_attn.v.bias', 'pipe.dit.blocks.7.self_attn.o.weight', 'pipe.dit.blocks.7.self_attn.o.bias', 'pipe.dit.block
3
+ warnings.warn(
4
+
5
+ ================================================================================
6
+ TRAINABLE PARAMETERS VERIFICATION
7
+ ================================================================================
8
+
9
+ Trainable modules (95 module groups):
10
+
11
+ 📦 pipe.dit.blocks.* — 40 blocks [0..39]:
12
+ Params per block: 131,107,840 × 40 = 5,244,313,600
13
+ Submodules (example from block 0):
14
+ • self_attn: 104,888,320
15
+ • camera_projector: 26,219,520
16
+
17
+ 📦 pipe.dense_target_control_encoder.conv3d_patchify:
18
+ Total params: 10,272
19
+ ├─ weight: 10,240
20
+ ├─ bias: 32
21
+
22
+ 📦 pipe.dense_target_control_encoder.null_token:
23
+ Total params: 156
24
+ pipe.dense_target_control_encoder.null_token: 156
25
+
26
+ 📦 pipe.dense_target_control_encoder.proj_up:
27
+ Total params: 5,148
28
+ ├─ weight: 4,992
29
+ ├─ bias: 156
30
+
31
+ 📦 pipe.dense_target_control_encoder.residual_blocks:
32
+ Total params: 110,720
33
+ ├─ 0.conv1.weight: 27,648
34
+ ├─ 0.conv2.weight: 27,648
35
+ ├─ 1.conv1.weight: 27,648
36
+ ├─ 1.conv2.weight: 27,648
37
+ ├─ 0.conv1.bias: 32
38
+ ├─ 0.conv2.bias: 32
39
+ ├─ 1.conv1.bias: 32
40
+ ├─ 1.conv2.bias: 32
41
+
42
+ 📦 pipe.dit._camera_patchify_conv:
43
+ Total params: 36,870
44
+ ├─ weight: 36,864
45
+ ├─ bias: 6
46
+
47
+ 📦 pipe.dit._camera_plucker_patchify_encoder:
48
+ Total params: 35,840
49
+ ├─ weight: 30,720
50
+ ├─ bias: 5,120
51
+
52
+ 📦 pipe.dit.control_patch_embedding:
53
+ Total params: 332,800
54
+ ├─ weight: 327,680
55
+ ├─ bias: 5,120
56
+
57
+ 📦 pipe.dit.domain_embedding:
58
+ Total params: 10,240
59
+ pipe.dit.domain_embedding: 10,240
60
+
61
+ 📦 pipe.track_adapter.conv3d_patchify:
62
+ Total params: 65,568
63
+ ├─ weight: 65,536
64
+ ├─ bias: 32
65
+
66
+ 📦 pipe.track_adapter.mlp:
67
+ Total params: 80,028
68
+ ├─ 0.weight: 30,888
69
+ ├─ 2.weight: 24,336
70
+ ├─ 4.weight: 24,336
71
+ ├─ 0.bias: 156
72
+ ├─ 2.bias: 156
73
+ ├─ 4.bias: 156
74
+
75
+ 📦 pipe.track_adapter.null_token:
76
+ Total params: 156
77
+ pipe.track_adapter.null_token: 156
78
+
79
+ 📦 pipe.track_adapter.proj_down:
80
+ Total params: 5,024
81
+ ├─ weight: 4,992
82
+ ├─ bias: 32
83
+
84
+ 📦 pipe.track_adapter.proj_up:
85
+ Total params: 5,148
86
+ ├─ weight: 4,992
87
+ ├─ bias: 156
88
+
89
+ 📦 pipe.track_adapter.residual_blocks:
90
+ Total params: 55,360
91
+ ├─ 0.conv1.weight: 27,648
92
+ ├─ 0.conv2.weight: 27,648
93
+ ├─ 0.conv1.bias: 32
94
+ ├─ 0.conv2.bias: 32
95
+
96
+ 📦 pipe.track_block_injector.encoders:
97
+ Total params: 1,080,729,600
98
+ ├─ 0.weight: 27,013,120
99
+ ├─ 1.weight: 27,013,120
100
+ ├─ 10.weight: 27,013,120
101
+ ├─ 11.weight: 27,013,120
102
+ ├─ 12.weight: 27,013,120
103
+ ├─ 13.weight: 27,013,120
104
+ ├─ 14.weight: 27,013,120
105
+ ├─ 15.weight: 27,013,120
106
+ ├─ 16.weight: 27,013,120
107
+ ├─ 17.weight: 27,013,120
108
+ ├─ 18.weight: 27,013,120
109
+ ├─ 19.weight: 27,013,120
110
+ ├─ 2.weight: 27,013,120
111
+ ├─ 20.weight: 27,013,120
112
+ ├─ 21.weight: 27,013,120
113
+ ├─ 22.weight: 27,013,120
114
+ ├─ 23.weight: 27,013,120
115
+ ├─ 24.weight: 27,013,120
116
+ ├─ 25.weight: 27,013,120
117
+ ├─ 26.weight: 27,013,120
118
+ ├─ 27.weight: 27,013,120
119
+ ├─ 28.weight: 27,013,120
120
+ ├─ 29.weight: 27,013,120
121
+ ├─ 3.weight: 27,013,120
122
+ ├─ 30.weight: 27,013,120
123
+ ├─ … (55 more)
124
+
125
+ ────────────────────────────────────────────────────────────────────────────────
126
+ NOTE: FSDP is active; counts below are from the rank-local post-wrap parameter view, not a full global parameter census.
127
+ Total trainable parameters: 6,325,796,530 (6325.80M)
128
+ Total frozen parameters: 10,363,571,200 (10363.57M)
129
+ Percentage trainable (overall): 37.90%
130
+ Percentage trainable (in requested modules: dit): 33.60%
131
+ ================================================================================
132
+
133
+ [torch-profiler] rank0 enabled (wait=3, warmup=1, active=5, repeat=1) -> /lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/torch_profiler
134
+ Epoch 1/200: 0%| | 29/7813 [08:21<37:24:54, 17.30s/it, opt_step=2559, loss=0.0673]
135
+ [Track4D-360] First train batch runtime flags: use_gradient_checkpointing=True, use_gradient_checkpointing_offload=False
136
+ [soft_splat_warp] WARN: tracks grid 160x294 (N=47040) != image grid 288x512 (N=147456); bilinear-resampling tracks_world to image grid. This warning fires once per process.
137
+ [track4d-runtime] rank0 patched DiTBlock.forward saw mode=track block_idx=0
138
+ [Attention] Using: Flash Attention 2 (flash_attn)
139
+
140
+ [Training] Reached max_train_steps=2560. Stopping training.
141
+
142
+ [Checkpoint] Saving 597 parameters (total 6,325,796,530 ≈ 6325.80M):
143
+
144
+ 📦 blocks.* — 40 blocks [0..39]:
145
+ Params per block: 131,107,840 × 40 = 5,244,313,600
146
+ Submodules (example from block 0):
147
+ • self_attn: 104,888,320
148
+ • camera_projector: 26,219,520
149
+
150
+ 📦 _camera_patchify_conv.bias:
151
+ Total params: 6
152
+ _camera_patchify_conv.bias: 6
153
+
154
+ 📦 _camera_patchify_conv.weight:
155
+ Total params: 36,864
156
+ _camera_patchify_conv.weight: 36,864
157
+
158
+ 📦 _camera_plucker_patchify_encoder.bias:
159
+ Total params: 5,120
160
+ _camera_plucker_patchify_encoder.bias: 5,120
161
+
162
+ 📦 _camera_plucker_patchify_encoder.weight:
163
+ Total params: 30,720
164
+ _camera_plucker_patchify_encoder.weight: 30,720
165
+
166
+ 📦 control_patch_embedding.bias:
167
+ Total params: 5,120
168
+ control_patch_embedding.bias: 5,120
169
+
170
+ 📦 control_patch_embedding.weight:
171
+ Total params: 327,680
172
+ control_patch_embedding.weight: 327,680
173
+
174
+ 📦 domain_embedding:
175
+ Total params: 10,240
176
+ domain_embedding: 10,240
177
+
178
+ 📦 pipe.dense_target_control_encoder.conv3d_patchify:
179
+ Total params: 10,272
180
+ ├─ weight: 10,240
181
+ ├─ bias: 32
182
+
183
+ 📦 pipe.dense_target_control_encoder.null_token:
184
+ Total params: 156
185
+ pipe.dense_target_control_encoder.null_token: 156
186
+
187
+ 📦 pipe.dense_target_control_encoder.proj_up:
188
+ Total params: 5,148
189
+ ├─ weight: 4,992
190
+ ├─ bias: 156
191
+
192
+ 📦 pipe.dense_target_control_encoder.residual_blocks:
193
+ Total params: 110,720
194
+ ├─ 0.conv1.weight: 27,648
195
+ ├─ 0.conv2.weight: 27,648
196
+ ├─ 1.conv1.weight: 27,648
197
+ ├─ 1.conv2.weight: 27,648
198
+ ├─ 0.conv1.bias: 32
199
+ ├─ 0.conv2.bias: 32
200
+ ├─ 1.conv1.bias: 32
201
+ ├─ 1.conv2.bias: 32
202
+
203
+ 📦 pipe.track_adapter.conv3d_patchify:
204
+ Total params: 65,568
205
+ ├─ weight: 65,536
206
+ ├─ bias: 32
207
+
208
+ 📦 pipe.track_adapter.mlp:
209
+ Total params: 80,028
210
+ ├─ 0.weight: 30,888
211
+ ├─ 2.weight: 24,336
212
+ ├─ 4.weight: 24,336
213
+ ├─ 0.bias: 156
214
+ ├─ 2.bias: 156
215
+ ├─ 4.bias: 156
216
+
217
+ 📦 pipe.track_adapter.null_token:
218
+ Total params: 156
219
+ pipe.track_adapter.null_token: 156
220
+
221
+ 📦 pipe.track_adapter.proj_down:
222
+ Total params: 5,024
223
+ ├─ weight: 4,992
224
+ ├─ bias: 32
225
+
226
+ 📦 pipe.track_adapter.proj_up:
227
+ Total params: 5,148
228
+ ├─ weight: 4,992
229
+ ├─ bias: 156
230
+
231
+ 📦 pipe.track_adapter.residual_blocks:
232
+ Total params: 55,360
233
+ ├─ 0.conv1.weight: 27,648
234
+ ├─ 0.conv2.weight: 27,648
235
+ ├─ 0.conv1.bias: 32
236
+ ├─ 0.conv2.bias: 32
237
+
238
+ 📦 pipe.track_block_injector.encoders:
239
+ Total params: 1,080,729,600
240
+ ├─ 0.weight: 27,013,120
241
+ ├─ 1.weight: 27,013,120
242
+ ├─ 10.weight: 27,013,120
243
+ ├─ 11.weight: 27,013,120
244
+ ├─ 12.weight: 27,013,120
245
+ ├─ 13.weight: 27,013,120
246
+ ├─ 14.weight: 27,013,120
247
+ ├─ 15.weight: 27,013,120
248
+ … (72 more)
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/run-20260602_091502-iwxrpu9h/files/requirements.txt ADDED
@@ -0,0 +1,208 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ videox-fun==1.0.1
2
+ track4d-360==0.1.0
3
+ func_timeout==4.3.5
4
+ click==8.3.2
5
+ pip==26.0.1
6
+ nvidia-cuda-nvrtc-cu12==12.6.77
7
+ ImageIO==2.37.3
8
+ numpy==1.26.4
9
+ boto3==1.42.91
10
+ albucore==0.0.24
11
+ lpips==0.1.4
12
+ torchsde==0.2.6
13
+ tomesd==0.1.3
14
+ yarl==1.23.0
15
+ typer==0.24.1
16
+ triton==3.5.0
17
+ cycler==0.12.1
18
+ multidict==6.7.1
19
+ OpenEXR==3.4.10
20
+ soxr==1.0.0
21
+ kiwisolver==1.5.0
22
+ simsimd==6.5.16
23
+ parso==0.8.6
24
+ einops==0.8.2
25
+ ftfy==6.3.1
26
+ platformdirs==4.9.6
27
+ groovy==0.1.2
28
+ nvidia-nvtx-cu12==12.6.77
29
+ sentry-sdk==2.58.0
30
+ certifi==2026.2.25
31
+ markdown-it-py==4.0.0
32
+ uvicorn==0.38.0
33
+ nvidia-cudnn-cu12==9.10.2.21
34
+ natsort==8.4.0
35
+ awscrt==0.31.2
36
+ accelerate==1.13.0
37
+ deepspeed==0.18.9
38
+ matplotlib==3.10.8
39
+ mdurl==0.1.2
40
+ regex==2026.4.4
41
+ nvidia-cuda-cupti-cu12==12.6.80
42
+ shellingham==1.5.4
43
+ joblib==1.5.3
44
+ zipp==3.23.1
45
+ ipython_pygments_lexers==1.1.1
46
+ executing==2.2.1
47
+ safetensors==0.7.0
48
+ typing-inspection==0.4.2
49
+ httpcore==1.0.9
50
+ semantic-version==2.10.0
51
+ wandb==0.26.0
52
+ asttokens==3.0.1
53
+ datasets==4.8.4
54
+ decorator==5.2.1
55
+ attrs==26.1.0
56
+ six==1.17.0
57
+ importlib_metadata==9.0.0
58
+ py-cpuinfo==9.0.0
59
+ requests==2.33.1
60
+ xxhash==3.6.0
61
+ pytorch-lightning==2.6.1
62
+ fsspec==2026.2.0
63
+ albumentations==2.0.8
64
+ soundfile==0.13.1
65
+ torchaudio==2.9.0
66
+ torchmetrics==1.9.0
67
+ nvidia-cusolver-cu12==11.7.1.2
68
+ scipy==1.17.1
69
+ jmespath==1.1.0
70
+ threadpoolctl==3.6.0
71
+ plyfile==1.1.2
72
+ peft==0.19.1
73
+ pycparser==3.0
74
+ Markdown==3.10.2
75
+ aiosignal==1.4.0
76
+ pydantic==2.13.2
77
+ decord2==3.3.0
78
+ safehttpx==0.1.7
79
+ scikit-image==0.26.0
80
+ fonttools==4.62.1
81
+ lazy-loader==0.5
82
+ smmap==5.0.3
83
+ starlette==1.0.0
84
+ Werkzeug==3.1.8
85
+ propcache==0.4.1
86
+ tensorboard==2.20.0
87
+ rich==15.0.0
88
+ modelscope==1.36.3
89
+ absl-py==2.4.0
90
+ traitlets==5.14.3
91
+ networkx==3.6.1
92
+ hf-xet==1.4.3
93
+ h5py==3.16.0
94
+ PyYAML==6.0.3
95
+ Jinja2==3.1.6
96
+ nvidia-cusparse-cu12==12.5.4.2
97
+ idna==3.11
98
+ transformers==4.46.2
99
+ httpx==0.28.1
100
+ MarkupSafe==3.0.3
101
+ tensorboard-data-server==0.7.2
102
+ pooch==1.9.0
103
+ torch==2.9.0+cu126
104
+ soupsieve==2.8.3
105
+ numba==0.65.0
106
+ torchdiffeq==0.2.5
107
+ jedi==0.19.2
108
+ charset-normalizer==3.4.7
109
+ annotated-doc==0.0.4
110
+ gradio==6.12.0
111
+ Pygments==2.20.0
112
+ pandas==3.0.2
113
+ stack-data==0.6.3
114
+ contourpy==1.3.3
115
+ tomlkit==0.14.0
116
+ scikit-learn==1.8.0
117
+ librosa==0.11.0
118
+ nvidia-cusparselt-cu12==0.7.1
119
+ pytz==2026.1.post1
120
+ go-with-the-track==0.1.0
121
+ yunchang==0.6.4
122
+ onnxruntime==1.24.4
123
+ ninja==1.13.0
124
+ sentencepiece==0.2.1
125
+ pexpect==4.9.0
126
+ multiprocess==0.70.19
127
+ omegaconf==2.3.0
128
+ h11==0.16.0
129
+ pydub==0.25.1
130
+ nvidia-cufft-cu12==11.3.0.4
131
+ xfuser==0.4.5
132
+ tifffile==2026.3.3
133
+ msgpack==1.1.2
134
+ psutil==7.2.2
135
+ ipython==9.12.0
136
+ hf-gradio==0.4.1
137
+ diffusers==0.33.1
138
+ python-dateutil==2.9.0.post0
139
+ orjson==3.11.8
140
+ huggingface_hub==0.36.2
141
+ nvidia-cuda-runtime-cu12==12.6.77
142
+ beautifulsoup4==4.14.3
143
+ ptyprocess==0.7.0
144
+ protobuf==7.34.1
145
+ GitPython==3.1.46
146
+ sympy==1.14.0
147
+ packaging==26.1
148
+ aiohttp==3.13.5
149
+ nvidia-nvshmem-cu12==3.3.20
150
+ annotated-types==0.7.0
151
+ antlr4-python3-runtime==4.9.3
152
+ gradio_client==2.4.1
153
+ stringzilla==4.6.0
154
+ flatbuffers==25.12.19
155
+ filelock==3.28.0
156
+ imageio-ffmpeg==0.6.0
157
+ fastapi==0.136.0
158
+ nvidia-nvjitlink-cu12==12.6.85
159
+ s3transfer==0.16.0
160
+ python-multipart==0.0.26
161
+ DistVAE==0.0.0b5
162
+ urllib3==2.6.3
163
+ nvidia-nccl-cu12==2.27.5
164
+ av==15.1.0
165
+ pure_eval==0.2.3
166
+ wheel==0.46.3
167
+ dill==0.4.1
168
+ flash_attn==2.8.3
169
+ track4d-360==0.1.0
170
+ opencv-python==4.11.0.86
171
+ anyio==4.13.0
172
+ pytest==9.0.3
173
+ nvidia-cufile-cu12==1.11.1.6
174
+ lightning-utilities==0.15.3
175
+ pluggy==1.6.0
176
+ grpcio==1.80.0
177
+ cffi==2.0.0
178
+ iniconfig==2.3.0
179
+ videox-fun==1.0.1
180
+ matplotlib-inline==0.2.1
181
+ tokenizers==0.20.3
182
+ pydantic_core==2.46.2
183
+ aiohappyeyeballs==2.6.1
184
+ typing_extensions==4.15.0
185
+ timm==1.0.26
186
+ pyarrow==23.0.1
187
+ mpmath==1.3.0
188
+ prompt_toolkit==3.0.52
189
+ mediapy==1.2.6
190
+ llvmlite==0.47.0
191
+ opencv-python-headless==4.11.0.86
192
+ hjson==3.1.0
193
+ tqdm==4.67.3
194
+ torchvision==0.24.0
195
+ pyparsing==3.3.2
196
+ nvidia-cublas-cu12==12.6.4.1
197
+ botocore==1.42.91
198
+ nanobind==2.9.2
199
+ wcwidth==0.6.0
200
+ gitdb==4.0.12
201
+ frozenlist==1.8.0
202
+ brotli==1.2.0
203
+ trampoline==0.1.2
204
+ pillow==12.2.0
205
+ audioread==3.1.0
206
+ lightning==2.6.1
207
+ setuptools==82.0.1
208
+ nvidia-curand-cu12==10.3.7.77
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/run-20260602_091502-iwxrpu9h/files/wandb-metadata.json ADDED
@@ -0,0 +1,235 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-6.4.0-150600.23.73_15.0.14-cray_shasta_c_64k-aarch64-with-glibc2.38",
3
+ "python": "CPython 3.12.13",
4
+ "startedAt": "2026-06-02T09:15:02.848946Z",
5
+ "args": [
6
+ "--model_id_with_origin_paths",
7
+ "Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/models_t5_umt5-xxl-enc-bf16.pth,Wan-AI/Wan2.1-T2V-14B:/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/Wan2.1_VAE.pth",
8
+ "--tokenizer_path",
9
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/google/umt5-xxl",
10
+ "--task",
11
+ "sft:v2v",
12
+ "--model_variant",
13
+ "vt2v",
14
+ "--v2v_concat_mode",
15
+ "width",
16
+ "--height",
17
+ "288",
18
+ "--width",
19
+ "512",
20
+ "--num_frames",
21
+ "49",
22
+ "--track4d360_data_config",
23
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/mixed_synth_recammaster_pair1x49_49f_288x512_ismb_nextstage.yaml",
24
+ "--steps_per_epoch",
25
+ "25000",
26
+ "--num_epochs",
27
+ "200",
28
+ "--learning_rate",
29
+ "2e-5",
30
+ "--gradient_accumulation_steps",
31
+ "1",
32
+ "--save_steps",
33
+ "500",
34
+ "--remove_prefix_in_ckpt",
35
+ "pipe.dit.",
36
+ "--trainable_models",
37
+ "dit.domain_embedding,dit.control_patch_embedding",
38
+ "--dit_finetune_modules",
39
+ "self_attn",
40
+ "--track_injection_mode",
41
+ "per_block",
42
+ "--track_injection_block_mode",
43
+ "concat_project",
44
+ "--track_config",
45
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/config/track4d_360/default_conv3d_patchify_srcdepth.yaml",
46
+ "--train_sigma_shift",
47
+ "5.0",
48
+ "--use_query_frame_impulse_condition",
49
+ "--use_plucker_camera_control",
50
+ "--enable_v2v_plucker_camera_control",
51
+ "--use_plucker_raymap_tokens",
52
+ "--use_perlayer_camera_control",
53
+ "--camera_injection_mode",
54
+ "projector",
55
+ "--camera_plucker_encoder_mode",
56
+ "patchify",
57
+ "--use_dense_branch",
58
+ "--dense_in_channels",
59
+ "5",
60
+ "--dense_proj_dim",
61
+ "32",
62
+ "--dense_num_residual_blocks",
63
+ "2",
64
+ "--dense_alpha_track",
65
+ "1.0",
66
+ "--dense_conditioning_noise_max_level",
67
+ "0.15",
68
+ "--source_mask_boundary_depth_dropout_prob",
69
+ "0.5",
70
+ "--source_mask_boundary_kernel_size",
71
+ "5",
72
+ "--dense_warp_mode",
73
+ "hybrid_softsplat",
74
+ "--dense_warp_motion_threshold",
75
+ "0.005",
76
+ "--cond_drop_geo_prob",
77
+ "0.1",
78
+ "--cond_drop_track_prob",
79
+ "0.1",
80
+ "--cond_drop_dense_prob",
81
+ "0.0",
82
+ "--initial_step",
83
+ "2530",
84
+ "--output_path",
85
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout",
86
+ "--dataset_num_workers",
87
+ "7",
88
+ "--dataloader_prefetch_factor",
89
+ "2",
90
+ "--drop_text_prob",
91
+ "0.1",
92
+ "--debug_latents_every",
93
+ "500",
94
+ "--allow_existing_checkpoints",
95
+ "--wandb_project",
96
+ "track4d-360",
97
+ "--wandb_name",
98
+ "ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp",
99
+ "--wandb_tags",
100
+ "ismb,track4d360,14b,per_block,conv3d,hybrid_dense,plucker_raymap_tokens,mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptoken,49f,288x512,concat_project,attn,cond_dropout,32gpu,8node,dense_noise_aug",
101
+ "--use_wandb",
102
+ "--vxf_init_checkpoint",
103
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/weights/Wan2.1-T2V-14B/diffusion_pytorch_model*.safetensors",
104
+ "--use_gradient_checkpointing",
105
+ "--resume_checkpoint",
106
+ "/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_cond_dropout/step-2530.safetensors",
107
+ "--max_train_steps",
108
+ "2560",
109
+ "--torch_profile_active_steps",
110
+ "5",
111
+ "--torch_profile_wait_steps",
112
+ "3",
113
+ "--torch_profile_warmup_steps",
114
+ "1",
115
+ "--torch_profile_repeat",
116
+ "1",
117
+ "--no_torch_profile_record_shapes",
118
+ "--no_torch_profile_memory"
119
+ ],
120
+ "program": "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360/scripts/wan2.1_fun/train_track4d360.py",
121
+ "codePath": "scripts/wan2.1_fun/train_track4d360.py",
122
+ "codePathLocal": "scripts/wan2.1_fun/train_track4d360.py",
123
+ "git": {
124
+ "remote": "git@github.com:NIRVANALAN/track4d-360.git",
125
+ "commit": "e47380b0c0f4cbb87efcbad540b96a404c2ccc88"
126
+ },
127
+ "email": "yushi.lan@outlook.com",
128
+ "root": "/lus/lfs1aip2/projects/public/u6ew/yslan_public/logs/track4d-360/ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout",
129
+ "host": "nid011121",
130
+ "executable": "/projects/public/u6ew/yslan_public/yslan_project/miniforge3/envs/th29py312/bin/python3.12",
131
+ "cpu_count": 288,
132
+ "cpu_count_logical": 288,
133
+ "gpu": "NVIDIA GH200 120GB",
134
+ "gpu_count": 4,
135
+ "disk": {
136
+ "/": {
137
+ "total": "255795724288",
138
+ "used": "549191680"
139
+ }
140
+ },
141
+ "memory": {
142
+ "total": "919613407232"
143
+ },
144
+ "gpu_nvidia": [
145
+ {
146
+ "name": "NVIDIA GH200 120GB",
147
+ "memoryTotal": "102625181696",
148
+ "cudaCores": 16896,
149
+ "architecture": "Hopper",
150
+ "uuid": "GPU-d655252a-055e-6123-1fcc-96c384e94cb6"
151
+ },
152
+ {
153
+ "name": "NVIDIA GH200 120GB",
154
+ "memoryTotal": "102625181696",
155
+ "cudaCores": 16896,
156
+ "architecture": "Hopper",
157
+ "uuid": "GPU-f32c462b-35e6-cd39-3ab1-ef2d12935745"
158
+ },
159
+ {
160
+ "name": "NVIDIA GH200 120GB",
161
+ "memoryTotal": "102625181696",
162
+ "cudaCores": 16896,
163
+ "architecture": "Hopper",
164
+ "uuid": "GPU-c6448e24-bbf3-db8b-e6c7-7becaf680d11"
165
+ },
166
+ {
167
+ "name": "NVIDIA GH200 120GB",
168
+ "memoryTotal": "102625181696",
169
+ "cudaCores": 16896,
170
+ "architecture": "Hopper",
171
+ "uuid": "GPU-2c148b70-ecf2-fb15-969b-cbb11a8b9bc4"
172
+ }
173
+ ],
174
+ "cudaVersion": "12.7",
175
+ "slurm": {
176
+ "cluster_name": "gracehopper",
177
+ "conf": "/var/spool/slurmd/conf-cache/slurm.conf",
178
+ "cpu_bind": "quiet,mask_cpu:0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF",
179
+ "cpu_bind_list": "0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF",
180
+ "cpu_bind_type": "mask_cpu:",
181
+ "cpu_bind_verbose": "quiet",
182
+ "cpus_on_node": "288",
183
+ "distribution": "cyclic",
184
+ "gpus": "32",
185
+ "gpus_on_node": "4",
186
+ "gpus_per_node": "4",
187
+ "gtids": "0",
188
+ "job_account": "brics.u6ew",
189
+ "job_cpus_per_node": "288(x8)",
190
+ "job_end_time": "1780398532",
191
+ "job_gid": "1483804847",
192
+ "job_gpus": "0,1,2,3",
193
+ "job_id": "4984787",
194
+ "job_name": "14b_pair1x49_profile_patchifylinear_fsdp2",
195
+ "job_nodelist": "nid[011121-011122,011125,011127-011129,011165,011167]",
196
+ "job_num_nodes": "8",
197
+ "job_partition": "workq",
198
+ "job_qos": "normal",
199
+ "job_start_time": "1780391332",
200
+ "job_uid": "1483804847",
201
+ "job_user": "yushi.u6ew",
202
+ "jobid": "4984787",
203
+ "launch_node_ipaddr": "10.242.4.57",
204
+ "localid": "0",
205
+ "mpi_type": "pmi2",
206
+ "nnodes": "8",
207
+ "nodeid": "0",
208
+ "nodelist": "nid[011121-011122,011125,011127-011129,011165,011167]",
209
+ "nprocs": "8",
210
+ "ntasks": "8",
211
+ "ntasks_per_node": "1",
212
+ "oom_kill_step": "0",
213
+ "prio_process": "0",
214
+ "procid": "0",
215
+ "srun_comm_host": "10.242.4.57",
216
+ "srun_comm_port": "45551",
217
+ "step_gpus": "0,1,2,3",
218
+ "step_id": "0",
219
+ "step_launcher_port": "45551",
220
+ "step_nodelist": "nid[011121-011122,011125,011127-011129,011165,011167]",
221
+ "step_num_nodes": "8",
222
+ "step_num_tasks": "8",
223
+ "step_resv_ports": "27631-27632",
224
+ "step_tasks_per_node": "1(x8)",
225
+ "stepid": "0",
226
+ "submit_dir": "/lus/lfs1aip2/projects/public/u6ew/yslan_public/yslan_project/track4d-360",
227
+ "submit_host": "login40",
228
+ "task_pid": "236308",
229
+ "tasks_per_node": "1(x8)",
230
+ "topology_addr": "global.group12.nid011121",
231
+ "topology_addr_pattern": "switch.switch.node",
232
+ "umask": "0022"
233
+ },
234
+ "writerId": "f0ztrk8t2qi0av8h7ptk8q57xoxkau1o"
235
+ }
ismb_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout_288x512x49_14b_32gpu_fsdp/train/Wan2.1-Fun-14B_track4d360_hybrid_dense_plucker_mixed_synth_recammaster_pair1x49_nextstage_softsplat_raymaptokens_eb32_288x512_profile_patchifylinear_cond_dropout/wandb/run-20260602_091502-iwxrpu9h/files/wandb-summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"train/track/injector/block_01/weight_norm":71.73619842529297,"train/timing/unit/WanVideoUnit_LatentNoiseAugmentation_s":5.2716633945237845e-05,"train/track/injector/block_00/grad_norm":0.03125328943133354,"train/timing/unit/WanVideoUnit_WarpedAppearance_s":5.737947503803298e-05,"train/track/injector/block_25/grad_norm":0.00310494564473629,"train/track/injector/block_29/weight_norm":71.70838165283203,"train/track/injector/block_16/weight_norm":71.75886535644531,"train/track/injector/block_08/grad_norm":0.009170296601951122,"train/track/injector/block_06/grad_norm":0.009117234498262405,"train/track/injector/block_11/grad_norm":0.00943054910749197,"train/track/injector/block_15/weight_norm":71.72622680664062,"_step":2559,"train/track/injector/block_33/weight_norm":71.6600341796875,"train/timing/optimizer_s":0.03643082920461893,"train/timing/unit/WanVideoUnit_PromptEmbedder_s":0.00025716342497617006,"train/track/injector/block_34/weight_norm":71.64431762695312,"train/track/injector/block_31/weight_norm":71.70207977294922,"train/track/injector/block_20/grad_norm":0.004566177260130644,"train/timing/microsteps":1,"train/track/injector/block_39/grad_norm":0.00405696127563715,"train/timing/unit/WanVideoUnit_NoiseInitializer_s":0.00019639797392301261,"train/track/injector/block_01/grad_norm":0.011140394024550915,"train/track/injector/block_10/weight_norm":71.77755737304688,"_timestamp":1.780392214034626e+09,"train/cond_dropout/drop_tracks":0.09375,"train/track/injector/block_35/weight_norm":71.68328094482422,"train/timing/data_wait_s":0.0015823920257389545,"train/timing/backward_s":5.116768166888505,"train/timing/reduction_s":0.14721127692610025,"train/track/injector/block_26/weight_norm":71.73145294189453,"train/timing/unit/WanVideoUnit_TrackPointCondition_s":0.019847922027111053,"train/timing/unit/WanVideoUnit_Vista4DStreams_s":5.4018557420931756e-05,"train/track/injector/block_27/weight_norm":71.70172882080078,"train/track/injector/block_04/grad_norm":0.009377378039062023,"train/hbm/post_t5_allocated_gb":28.869243621826172,"train/track/injector/block_28/weight_norm":71.74332427978516,"train/track/injector/block_21/grad_norm":0.00487802317366004,"train/timing/step_total_s":11.40470616798848,"train/track/injector/block_32/grad_norm":0.0020946995355188847,"train/track/injector/block_31/grad_norm":0.0020167483016848564,"train/track/injector/block_02/grad_norm":0.010800544172525406,"train/track/injector/block_14/weight_norm":71.7539291381836,"train/track/injector/block_17/weight_norm":71.72591400146484,"train/track/injector/block_37/grad_norm":0.0018434051889926195,"train/track/injector/block_04/weight_norm":71.78131103515625,"train/timing/unit/WanVideoUnit_InputVideoEmbedder_XYZCondition_s":1.6512749195098877,"train/track/injector/block_36/grad_norm":0.0018603148637339473,"_wandb":{"runtime":526},"train/track/injector/block_19/weight_norm":71.72638702392578,"train/timing/unit/WanVideoUnit_TeaCache_s":2.1591491531580687e-05,"train/track/injector/block_03/grad_norm":0.010950344614684582,"train/track/injector/block_28/grad_norm":0.0027651339769363403,"train/track/injector/block_09/grad_norm":0.009453365579247475,"train/track/injector/block_05/grad_norm":0.009369931183755398,"train/track/injector/block_13/weight_norm":71.73658752441406,"train/timing/unit/WanVideoUnit_CameraControl_NVS_s":0.003380503272637725,"train/track/injector/block_26/grad_norm":0.002930134069174528,"train/track/injector/block_25/weight_norm":71.7306137084961,"train/track/injector/block_22/weight_norm":71.7159423828125,"train/timing/forward_s":5.396990212611854,"train/track/injector/block_14/grad_norm":0.008379348553717136,"train/t5/on_cuda":-1,"train/cond_dropout/drop_dense":0.03125,"train/track/injector/block_30/weight_norm":71.70831298828125,"train/track/injector/block_17/grad_norm":0.0065108006820082664,"train/track/injector/block_08/weight_norm":71.77717590332031,"train/epoch":0,"train/track/injector/block_02/weight_norm":71.7703857421875,"train/timing/dit_loss_fn_s":2.527306079864502,"train/track/injector/block_29/grad_norm":0.0027308855205774307,"train/track/injector/block_03/weight_norm":71.80191802978516,"train/track/injector/block_20/weight_norm":71.71338653564453,"train/hbm/reserved_gb":39.55208969116211,"train/track/injector/block_30/grad_norm":0.002292318269610405,"train/track/injector/block_18/grad_norm":0.006294068414717913,"train/track/injector/block_16/grad_norm":0.007425781339406967,"train/track/injector/block_00/weight_norm":71.62019348144531,"train/track/injector/block_11/weight_norm":71.75846862792969,"train/timing/unit/WanVideoUnit_ModeDropoutAugmentation_s":3.1350209610536695e-05,"train/track/injector/block_05/weight_norm":71.79116821289062,"train/track/injector/block_19/grad_norm":0.005807674024254084,"train/track/injector/weight_norm_max":71.80191802978516,"train/track/injector/block_06/weight_norm":71.78491973876953,"train/track/injector/block_07/grad_norm":0.00951184332370758,"train/track/injector/block_21/weight_norm":71.71271514892578,"train/timing/unit/WanVideoUnit_UnifiedSequenceParallel_s":2.87372458842583e-05,"train/track/injector/block_22/grad_norm":0.005098571069538593,"train/timing/unit/WanVideoUnit_TargetVideoEncoder_s":0.829370379447937,"train/track/injector/block_39/weight_norm":71.61335754394531,"train/loss":0.06727957725524902,"train/timing/unit/WanVideoUnit_DenseTargetControl_s":0.25093698501586914,"_runtime":526,"train/track/injector/block_27/grad_norm":0.0027274005115032196,"train/timing/other_s":0.7073056823574007,"train/track/injector/grad_norm_max":0.03125328943133354,"train/hbm/allocated_gb":3.4353647232055664,"train/track/injector/block_07/weight_norm":71.78191375732422,"train/track/injector/block_36/weight_norm":71.64971923828125,"train/track/injector/block_10/grad_norm":0.009506093338131905,"train/track/injector/block_37/weight_norm":71.64720916748047,"train/track/injector/block_23/grad_norm":0.004205186385661364,"train/timing/unit/WanVideoUnit_V2VConcatenate_s":0.0001357975124847144,"train/track/injector/block_32/weight_norm":71.66665649414062,"train/track/adapter/weight_norm":17.002626419067383,"train/track/injector/block_24/weight_norm":71.72339630126953,"train/track/injector/block_23/weight_norm":71.7166976928711,"train/track/adapter/grad_norm":0.0050230869092047215,"train/track/injector/block_38/weight_norm":71.60942077636719,"train/track/injector/block_34/grad_norm":0.0019225106807425618,"train/track/injector/weight_norm_mean":71.7211685180664,"train/track/injector/block_12/grad_norm":0.008436022326350212,"train/track/injector/block_13/grad_norm":0.007800602354109287,"train/track/injector/block_33/grad_norm":0.0021519330330193043,"train/timing/unit/WanVideoUnit_ShapeChecker_s":5.8914534747600555e-05,"train/hbm/post_t5_reserved_gb":28.94489097595215,"train/track/injector/grad_norm_mean":0.006389424670487642,"train/track/injector/block_38/grad_norm":0.0021260089706629515,"train/track/injector/block_24/grad_norm":0.00367747456766665,"train/track/injector/block_12/weight_norm":71.7568359375,"train/track/injector/block_09/weight_norm":71.78079223632812,"train/track/injector/block_35/grad_norm":0.0019217939116060734,"train/track/injector/block_15/grad_norm":0.00687079643830657,"train/timing/unit/WanVideoUnit_CfgMerger_s":1.3017634046263993e-05,"train/track/injector/block_18/weight_norm":71.7179183959961}